[0.3.248] - 2026-11-18
0.3.247 was tagged but never published: its release check found that mockforge cve failed on Linux, so 0.3.248 ships its changes with the fix.
Security
- [DevX] Update
libz-systo 1.1.33 for RUSTSEC-2026-0544 (heap buffer overflow). (#1630) - [DevX] Update
lettreto 0.11.24 for six SMTP advisories, RUSTSEC-2026-0560 to RUSTSEC-2026-0565:- STARTTLS response injection that lets a man-in-the-middle forge SMTP replies;
- SMTP command injection through missing dot-stuffing;
- mail for a quoted local part sent to a different address;
- unbounded reply buffering;
- two panics (denial of service). This affects outbound mail from the SMTP mock, reporting, pipelines and the registry. (#1648)
Fixed
- [Reality]
mockforge cveworks again. It passed every bundled CVE lab module to Python as one command-line argument, and once the bundle grew past Linux’s 128 KiB per-argument limit, everymockforge cvecommand failed with “Argument list too long”, reported as needing Python 3.11+. The modules now go to a private temporary directory, and a missingpython3is reported only when it is really missing. (#1654) - [Reality] CVE lab
inconclusiveresults explain themselves. When an unverified pack’s lab starts but its baseline does not reproduce, the reason says this is a known state of the pack, not a problem with your target or setup, and names the verified packs that run end to end. When your target blocks the exploit with a status other than--block-status, the reason names the status it returned and suggests--block-status <code>.report.jsongainspack_verified. (#1645, reported on #79) - [Reality] The managed CVE lab guide states each pack’s real status, explains what the lab sends to your target (not a replay of the baseline), and explains that pass or fail is judged by what reaches the lab origin through your load balancer. It also says where an extra test case goes. (#1645, #1646, #1611)
- [Reality] CVE labs deliver the generated Nuclei template into the runner container with create and copy instead of a host bind mount, so runs work when the engine cannot see the host path (Docker Desktop, podman machine and other remote or VM engines). (#1616)
- [Reality] A CVE lab pack without a
verifiedfield is treated as unverified. (#1599) - [Reality] A failed main signature is never softened by signature-variant evidence: a run whose main signature fails reports
faileven when a variant is inconclusive. (#1604) - [DevX] The
fuzz_openapi_parserfuzz target compiles again, and all nine fuzz targets build. Three stale member lockfiles that pinned vulnerable dependencies are removed, and the fuzz lockfile is refreshed. (#1634, #1632) - [DevX] CI fixes:
- the browser SDK audit covers shipped dependencies only (#1636);
- the scheduled hosted-mocks smoke test skips with a notice while its secrets are unset (#1629);
- the load tests serve only HTTP, admin and metrics (#1617);
- the coverage job uses the shared build target directory (#1614);
- the image smoke test’s egress check uses its own redirector and counts blocked requests correctly (#1640).
- the production smoke test runs on GitHub-hosted runners, so it no longer queues behind builds on the shared runners (#1652).
- [DevX] Documentation copy fixes: the FAQ pricing claim, the canonical description and the protocol list. (#1565)
Added
- [Reality] Every CVE lab run writes
sent_requests.jsonlnext toreport.json. It holds one line per request the lab sends to its own origin or through--target: the correlation marker, method, path, redacted headers, a body hash and preview, and the response status or error. Joined withreport.json’sorigin_observations, it shows what was sent versus what arrived through your load balancer. The guide addstcpdump/tsharkcommands for a byte-level capture. (#1647, reported on #79) - [Reality] Node2Shell CVE-2026-111497 pack (Node.js Brotli decoder out-of-bounds write) with an HTTP/2-capable recorder. It is unverified, because its pinned image has not yet reproduced the crash on a lab host, so it runs only with
--include-unverified. (#1523) - [Reality] Signature variants for the CVE-2026-109619
ini_set,php_valueandphp_admincases. (#1603)
[0.3.246] - 2026-11-15
Security
- [Cloud] The registry’s showcase admin endpoints require a site administrator. (#1590)
- [DevX] Update
quinn-prototo 0.11.19 andquinn-udpto 0.5.16 for RUSTSEC-2026-0523, RUSTSEC-2026-0524 and RUSTSEC-2026-0538. (#1580)
Fixed
- [Reality]
mockforge cve run --skip-tls-verifyagainst an HTTPS target no longer fails at the baseline withflag provided but not defined: -insecure. The pinned Nuclei 3.11.1 runner has no such flag and never verifies TLS certificates, so MockForge no longer passes one;--skip-tls-verifystill governs MockForge’s own control requests and is recorded in the report. A regression test runs the real pinned runner against a self-signed HTTPS server. (#1601, reported on #79) - [Cloud] The registry acknowledges Stripe webhooks that can never be acted on (no valid org, an unknown org, or an already-passed trial end) instead of answering a retryable 503 that Stripe would retry for days. (#1575)
Added
- [Reality] CVE lab packs can declare more test cases as data (
extra_cases), and must-be-blocked signature variants that run in the direct baseline and through the load balancer with a verdict per variant. (#1541, #1597) - [Reality]
mockforge cve listsurfaces each pack’s digest-pinned images and whether this machine holds them; digest-pulled images do not appear indocker images. (#1542) - [Reality] PHP2Shell 2026 packs for CVE-2026-109618 and CVE-2026-109619, with signature variants for the
PHP_SESSION_UPLOAD_PROGRESSplacements reported on #79. Both packs are unverified (their trigger has not reproduced on a lab host yet), so they run only with--include-unverified. (#1524, #1597) - [DevX] CI reclaims oversized runner build targets, with a 25G default cap. (#1581, #1586)
[0.3.245] - 2026-11-13
0.3.244 was tagged but never published (its release build ran out of disk), so 0.3.245 ships its changes.
Fixed
- [Contracts] Generated mock responses honour the schema’s constraints: string
enum,formatand length limits; integer and numberenum, bounds (including exclusive bounds) andmultipleOf; and booleanenum. This applies inside objects, arrays,$refandallOf/oneOf/anyOf. Constrained ids no longer get the_001suffix. Unconstrained values are unchanged. (#1558) - [Reality] When a CVE lab’s Nuclei step fails, the report and the printed reason now show the failing phase, the container’s exit code and the end of its output, instead of an empty
docker command failed:. (#1529) - [DevX] Releases no longer rebuild the workspace from scratch for the
cargo installsmoke test, and trim the runner’s release build directory afterwards, so a release stops filling the shared CI disk. (#1570) - [DevX] CI fixes: the CLI tools bundle is regenerated, the example configuration image workflow uses the standard runner selection, and two unneeded qualifications in
mockforge-proxytests are removed.
[0.3.243] - 2026-11-12
Fixed
- [DevX] The release check for bundled tools reads
mockforge cve listas one entry per pack, keyed by CVE id. v0.3.242 was tagged but never published because this check still expected the single-pack shape.
Added
- [DevX] A reproducible comparison benchmark of MockForge, WireMock, Prism and Mockoon (
benchmarks/compare/), with its methodology in the book. Results will be published after a rerun. (#1512)
v0.3.243 also carries everything listed for 0.3.242 below, which was never published to crates.io.
[0.3.242] - 2026-11-11
Security
- [DevX] Update
opensslto 0.10.82 (openssl-sys0.9.118) for eight RustSec advisories, including buffer overflows in AES key wrap and digest finalization. RUSTSEC-2026-0459 (libsqlite3-sys) is deferred to the sqlx 0.9 upgrade (#1552); mockforge only opens its own SQLite stores.
Added
- [Reality] CVE lab pack for CVE-2024-4577 (PHP-CGI argument injection through the Windows best-fit code page), alongside the Next.js pack. (#1522)
Fixed
- [DevX] UI builds that change nothing no longer regenerate watched assets. (#1488)
- [DevX] The UI restart helper builds cleanly with
clippy -D warningson non-Linux targets. (#1517)
Changed
- [DevX] Copyright holder, crate authors and publisher are now MockForge LLC (licenses unchanged). (#1528)
- [Reality] The CVE labs guide covers offline behavior, the plain-HTTP recorder, plain-language checks and the test cases each pack runs. (#1530)
- [DevX] Removed dead code and orphaned end-to-end test files, and brought the docs’ crate list in line with the workspace’s 63 crates. (#1505, #1513, #1518)
[0.3.241] - 2026-11-07
Added
- [Reality] CVE lab packs are now data-driven: each pack declares its catalog entry and its own evidence checks, and the lab recorder can forward POST bodies, which body-injection reproductions need.
- [Reality]
mockforge cve runstates its conclusion plainly:<target> is EXPOSED to <CVE>or<target> is NOT exposed to <CVE>, followed by the reason and each evidence check. The checks are also stored inreport.json. - [DevX]
mockforge cve runprints its plan before starting (recorder endpoint, target under test, and a warning when the default 127.0.0.1 bind can’t accept a remote load balancer).mockforge cve listshows each pack’sorigin_portand listen-address guidance. - [DevX]
mockforge cve run --skip-tls-verifyfor self-signed lab targets, mirroringmockforge bench. Verification stays on by default; the flag covers the control requests and the test client, warns while active, and is recorded in the report. - [DevX] The managed CVE labs guide has an “Air-gapped lab hosts” section:
docker saveon a connected machine,docker loadon the lab host, thenprepareverifies the pinned digests without pulling.
Fixed
- [Cloud] Tunnels refuse unauthenticated cloud destinations.
- [DevX] Bundled demo samples serve their real image responses.
- [Cloud] Stripe subscription webhooks reconcile ordered subscription state and handle API-version changes, repeats, unknown events and prices.
Changed
- [DevX] Protocol support is split out of transport builds, so builds that don’t need every protocol compile less.
[0.3.240] - 2026-11-03
Changed
- [DevX] Extract authentication middleware, token lifecycle and OAuth2/OIDC endpoints into
mockforge-auth. Preserve HTTP authentication and OAuth2 handler import paths and shared state types; keep the risk-simulation implementation with its engine. (#635)
Added
-
[Contracts][DevX]
mockforge validate-spec <spec> [--format json]validates one local OpenAPI 2.0 or 3.x document with the validatorgenerateuses, without generating, starting, fetching or writing anything. The JSON form is a versioned document (schema_version1) listing each finding’s severity, rule, JSON pointer, message and suggestion, for CI and agent hosts such as Aether. Exits 0 when valid, 1 with errors, 2 when the file is not an OpenAPI document it could read. -
[Reality] Add
mockforge cvemanaged labs with pinned vulnerable applications and Nuclei containers. The first pack reproduces Next.js CVE-2025-29927, checks direct and load-balancer paths, and reports blocking only when healthy controls and correlated origin evidence support it. (#79) -
[DevX] Automatically fetch managed CVE lab images, export evidence, and clean up run-owned containers. Include environment diagnostics and recovery commands for local Docker or Podman on Linux amd64 with Python 3.11+. (#79)
-
[Cloud] Share Free hosted capacity, monthly request limits and platform AI credits across a verified owner’s organizations. Preserve spent allowances through account recreation, bound deleted identity retention, retain uncertain provider budgets, and require audited usage evidence plus execution fences for stopped-call recovery. Keep provider execution connections separate from API requests. (#834, #868) The Privacy Policy and DPA disclose OpenAI as the platform AI sub-processor, with processing beginning no earlier than November 14, 2026.
-
[Reality] AsyncAPI 2/3 fidelity compares channel and directional operation contracts, including payload/header schemas, bindings, server/security references and traits. CLI and authenticated admin comparisons can check recorder message samples; the CLI also checks complete recorder-JSON fixture sets. Recorded MQTT, Kafka, AMQP and WebSocket messages use protocol-aware direction, exact JSON/text schema validation and explicit unverified outcomes when capture or format support is incomplete. Capture reads remain read-only and filter message events before the limit. (#849)
-
[Reality] The self-hosted Admin UI adds a Mock fidelity panel in Contract Diff for comparing mock/source OpenAPI definitions and optional Flight Recorder HTTP samples. Authenticated comparisons report per-route definition deltas, sample drift, and incomplete coverage, with bounded inputs and paginated details. (#849)
-
[Reality]
verify-mocks --fixtureschecks a complete HTTP replay fixture set against its source OpenAPI definition, reporting missing and removed routes, invalid statuses, and field-level JSON response errors. It can run alongside mock-definition and recorder checks. Loading is read-only and rejects empty, malformed, symlinked or excessive fixture inputs. (#849) -
[Reality]
verify-mocks --mock-speccompares the OpenAPI definition used by a mock with its current source specification. Structured reports include added, removed and changed endpoints, field-level JSON-pointer deltas, and per-route fidelity status. Definition checks can run alone or alongside captured upstream traffic and use the existing--fail-on-driftCI gate. (#849)
Fixed
-
[DevX] Identify configured mock-response routes in the control-plane mount audit. Keep generic router mounts rejected and management authentication checks intact. The unpublished 0.3.239 candidate failed this audit.
-
[DevX] Update the locked Markdown parser to pulldown-cmark 0.13.6 for RUSTSEC-2026-0433. Document the unreachable network-input path underlying the narrowly scoped Thrift RUSTSEC-2026-0435 audit exception.
-
[Auth] Authenticated handlers can extract
AuthClaimswith Axum’s standard extractor; middleware stores raw claims and the legacy wrapper while preserving credential-safe security events. (#635) -
[Security] Admin database upgrades preserve the shipped SQLx migration checksum and remove only unchanged built-in accounts in a new migration. Operator credentials and unrelated sessions are retained. (#1016)
-
[AI] Use the selected provider/model for direct and chat debugging instead of a default cloud client. Apply workspace validation and seed overrides to the debug endpoints, including when runtime MockAI is disabled.
-
[AI] Keep the chosen provider/model and seed when applying Reality presets. Baked fixtures disable presets and reject cached configurations that would re-enable them. (#852)
-
[AI] Retain the configured authoring provider/model when runtime MockAI is disabled. Pass trusted startup defaults separately so authenticated authoring and snapshot replay work without enabling AI for mock traffic. (#852)
-
[AI] Forward persisted workspace seed and snapshot policies through local admin chat, mock and persona generation. Inherit the server provider/model, preserve policy on export/import, and reject missing or invalid workspaces before generation. (#852)
-
[AI] Keep configured provider/model and seed settings when chat generates mocks or personas. Allow library callers to supply snapshot policies and report zero new usage for verified snapshot replays. (#852)
-
[AI] Preserve frozen array and scalar payloads with durable identity and integrity metadata. Publish immutable snapshots as complete bundles, retain legacy object snapshots, and reject missing hashes during verification. (#852)
-
[AI] Add the unreleased
mockforge ai bakeauthoring workflow: generate a JSON response once and save a static JSON/YAML serve configuration for offline CI. Existing verified outputs are reused without provider calls; mismatched inputs and modified files fail instead of overwriting them. Sampling seeds are forwarded where supported. Explicit JSONnullroute bodies now remainnullinstead of becoming{}when configuration loads. (#852) -
[Cloud] Path-based hosted-mock URLs work below the mock’s root.
/mocks/{org}/{slug}/<path>returned HTTP 500 “Wrong number of path arguments” for every sub-path, so only/mocks/{org}/{slug}itself was reachable. Sub-paths and query strings now proxy to the deployment, and a trailing slash (/mocks/{org}/{slug}/) serves the mock’s root instead of 404. (#1280) -
[Security] Require verified email-domain ownership for every SAML and OIDC login, including existing organization members, before resolving a global user identity.
-
[Reality] Fidelity checks preserve AsyncAPI server variables whose names match documentation keys. HTTP captures and replay fixtures now report unsupported or ambiguous response media as unverified, and
--fail-on-driftfails those checks instead of accepting unchecked responses. Recorded verification and project status use schema 2 with anunverifiedoutcome and continue reading schema-1 records. (#849) -
[Reality] JSON-content request parameters are decoded and checked against their schemas instead of silently passing validation. JSON types stay distinct, local component references are resolved, malformed JSON fails even without a schema, and unsupported parameter media types fail explicitly. (#849)
-
[Reality] Request parameter
schemareferences are checked in normal, style-aware and detailed validation instead of silently passing. Local aliases, nested and recursive components retain their schema context; referenced arrays and deep-object values are coerced by declared type, and integer parameters retain 64-bit precision. Unresolved/cyclic references fail explicitly, and parameter validation does not fetch external resources. (#849) -
[Reality] Request conformance checks match HTTP header names without ASCII case differences in normal and detailed validation. A captured
x-tracenow satisfies a declaredX-Trace; missing headers and invalid values still fail, while cookie, query and path parameter names remain case-sensitive. (#849) -
[Reality]
verify-mocksreads declared cookie parameters from recordedCookieheaders using serving’s case and delimiter rules. Previously, even a valid captured cookie was treated as missing. Missing or invalid required cookies still report request drift. (#849) -
[Reality] Fidelity checks treat omitted or empty root server lists as the OpenAPI default
/, preventing false drift between equivalent Swagger/OpenAPI definitions while retaining real server changes and route overrides. Recorded CLI query parameters now use the serving parser, including form spaces (+), encoded literal plus signs and empty values. (#849) -
[Reality]
verify-mocksenforces captured request contracts even when server validation is set toofforwarn, route overrides disable checking, or shadow mode is enabled. The offline verifier uses an isolated strict registry policy; serving retains its configured behavior. (#849) -
[Reality] Fidelity checks prefer literal path segments over parameters, matching HTTP routing. Previously, registry iteration order could select a parameter route’s schema for a literal endpoint, randomly passing invalid responses or failing valid ones. (#849)
-
[Reality]
verify-mocks --fail-on-driftnow fails on unknown endpoints, missing responses and undeclared statuses, and validates range/default/referenced response schemas. Empty captures still succeed (and--recordstores ano_trafficverification), databases open read-only, and the positive--limitapplies to recent HTTP records in SQL rather than silently stopping at 100 or being displaced by other protocols. (#849)
[0.3.238] - 2026-10-30
Added
-
[DevX]
mockforge tools export --output-dir <new-directory>exports the discussion exporter, CVE tools, Python modules, and Next.js example bundled in the crates.io CLI. A source checkout is no longer required. (#79) -
[DevX] CVE monitor and remediation tips for the rule packs:
cve_pack_monitor.pytracks CISA KEV (actively-exploited CVEs), triages to WAF-actionable CWEs, and writes a candidate pack per CVE enriched with tips aggregated and deduped from NVD, OSV, EPSS and KEV (upgrade-to version, exploitation probability, ransomware use, CISA due date);cve_pack.py tips <CVE>prints the same. An optional provider-agnostic--drafter-cmdproposes each signature, but the profile schema is the trust boundary — a draft that is not inert marker-only is flagged for review, never run. (#79) -
[DevX] Per-CVE Rule Packs: author a small profile (route, signature header/param names with inert marker values, benign controls, expected block status) and
cve_pack.pyscaffolds it, emits a MockForge mock of the guarded route, and runs the signature plus controls through a supplied load balancer to assert the rule blocks the signature and allows benign traffic. Reuses the Security Rule Test Packs transport; marker-only values mean a profile cannot carry a payload, and runtime/patch evidence stays unverified. (#79) -
[DevX] Preview Security Rule Test Packs with bounded safe Next.js/Camel transport canaries, benign controls, explicit status assertions and correlated origin reports. Runtime/patch evidence remains unverified; mixed-case Camel headers are unsupported. (#79)
-
[DevX] Per-CVE Rule Packs: author a small profile (route, signature header/param names with inert marker values, benign controls, expected block status) and
cve_pack.pyscaffolds it, emits a MockForge mock of the guarded route, and runs the signature plus controls through a supplied load balancer to assert the rule blocks the signature and allows benign traffic. Reuses the Security Rule Test Packs transport; marker-only values mean a profile cannot carry a payload, and runtime/patch evidence stays unverified. (#79) -
[DevX] Preview Security Rule Test Packs with bounded safe Next.js/Camel transport canaries, benign controls, explicit status assertions and correlated origin reports. Runtime/patch evidence remains unverified; mixed-case Camel headers are unsupported. (#79)
-
[DevX] Preview Security Rule Test Packs with bounded safe Next.js/Camel transport canaries, benign controls, explicit status assertions and correlated origin reports. Runtime/patch evidence remains unverified; mixed-case Camel headers are unsupported. (#79)
-
[DevX] Add a paginated issue-discussion exporter (searchable offline HTML, Markdown and JSON), a vendor-referenced CVE traffic lab guide, and bounded Next.js/Camel transport canaries with a real-client/server smoke test. The examples establish request delivery, not exploit reproduction. (#79)
-
[DevX]
mockforge status [--dir <path>] [--format json]reports a project’s MockForge state without starting or writing anything: the configuration fileservewould use, the HTTP port and spec it declares, whether a MockForge server answers there (through the management API, which a spec’s own/healthcannot shadow), and the last recordedverify-mocksresult. The JSON form is a versioned document (schema_version1) for editors and agent hosts such as Aether.verify-mocks --recordkeeps a run’s result in.mockforge/reports/verify-mocks.jsonfor it. (#1420) -
[Cloud] Trial owners get an email three days before their Pro or Team trial ends. The registry now handles Stripe’s
customer.subscription.trial_will_endwebhook and sends the reminder through the same email path as the other billing emails, honoring the owner’s email-notification setting. It gives the end date and price, and says the plan continues if a payment method is on file and otherwise drops to Free. (#714) -
[AI] Add a feature-gated Qdrant RAG storage backend with persistent chunk upserts, server-side filtered vector search, paged document reads, native snapshots, health checks and real-server restart acceptance. Enable
mockforge-data/qdrantto useStorageFactory::create_vector_db. (#717)
Changed
- [DevX]
bench-chunkedlabels each--specoperation’sRPS (started)andRPS (overall)lines(this operation only), so they are not read as the rate for the whole pass. The multi-target summary now printsload_window=andelapsed=next tostart_rpsandrps, and a single--targetrun with--specprints the same figures on aPass summary (all operations):line. The campaign banner says that--repeat-untilis checked before each pass, so a running pass finishes. (#79) - [Cloud] Hosted mocks that use HTTP, WebSocket or GraphQL can run on Cloudflare Containers at
https://mocks.mockforge.dev/<org>/<slug>. Each mock’s container starts on its first request and sleeps after 10 minutes idle, so an unused mock costs nothing. The Worker applies the plan’s body and requests-per-second caps and reports usage to the registry, which meters it as before. Enabled by settingMOCKFORGE_CF_MOCKS_URLandMOCKFORGE_CF_MOCKS_EDGE_TOKEN; other protocols and plugin-enabled mocks stay on Fly. Fly machines now stop when idle and start on the next connection. Redeploys use the same machine environment and protocol services as the first deploy (they used to drop protocol, log-ingest, tracing and upstream settings), deleting a running Fly machine no longer fails with 412, and the registry’s/mocks/<org>/<slug>redirects Cloudflare-served mocks to their edge URL. (#1063) - [UI] Marketplace navigation and hosted workspace/token controls use the shared Tailwind/CVA controls and Lucide icons instead of MUI compatibility components. Marketplace routes expose native links and current-page state, workspace selection keeps explicit assignment, and control-token guidance covers management requests. (#722)
Fixed
-
[Security] Admin SSE connections check the browser-resolved origin before attaching session credentials, including uppercase schemes, whitespace, and backslash URL forms.
-
[DevX] Discussion exports work without GitHub CLI; CVE tips preserve partial feed results and check CISA KEV; unfinished CVE templates are rejected before tests run. (#79)
-
[DevX] Add a guarded operator for superseded, partially cancelled CI runs stuck on an unassigned completion job. It preserves the fail-closed Rust completion verdict, tries ordinary cancellation first, and revalidates the live PR head and all job guards before force-cancellation. Current-head runs, forks and active workers are refused. (#1191)
-
[Cloud] Cloud test runs no longer report
passedwithout doing the work. Every synthetic path in the test runner is gone. Kinds with no real executor (unit,behavioral_clone,snapshot_capture,snapshot_restore,verification_suite,scenario,orchestration,state_machine) are rejected with a 400 when a suite is created or a run is triggered, and the scheduler skips them. Behavioral clone training is shown as not available yet, and onlychainflows can be created. Missing configuration (notarget_url, no steps, no links, no spec, nouse_cloud_apifor conformance) and zero executed checks reporterroredwith the reason, and errored runs bill 0 runner seconds. Previously these reportedpassed, which also passedmockforge cloud test run --waitin CI. Existing suites of removed kinds stay readable and editable. Chaos campaigns now need at least one declared fault.mockforge cloud testand the run page show the runner’s error for a run that did not complete. (#1264, #1270) -
[Cloud] Cloud OWASP runs write a k6
summary.jsonand fail when the scan reports findings. A rejected attack probe (4xx) no longer counts as a failed request, so a clean scan passes, and a target that never answers below 500 reportserrored. (#1264, #1270) -
[Contracts] Monitored services registered with only an inline OpenAPI spec are now diffed against it on manual verify and scheduled probes; a service with no spec gets a 400 on verify and is skipped by the probe worker. (#1264, #1270)
-
[Reality] MQTT publish and last-will delivery no longer hold broker-wide locks while waiting on subscriber queues, or acquire topic and client locks in the opposite order to new connections. This prevents concurrent CONNECT/publish deadlocks and lets new clients connect while a subscriber is backpressured. (#715)
-
[DevX] The editor and browser clients build and test again, and CI now runs them (
Client packagesworkflow). The JetBrains plugin never compiled against its target IDE and could not have loaded: it depended on a TOML plugin ID that does not exist, its config file type had no language, and its scenario generator was registered as a menu action instead of an intention. It now builds, passes./gradlew test buildPlugin, and ships a Gradle wrapper. The VS Code extension has the test runnernpm testpointed at, with tests that activate it in VS Code 1.85.2 and check every contributed command is registered. The browser extension type-checks clean, and it now captures request headers: they were read fromonBeforeRequest, which never includes them. (#1282) -
[Docs] The workspace export/import guide and the configuration reference no longer call
mockforge workspace export,importandactivateunreleased; they ship in 0.3.236. (#1197) -
[Contracts] Package Intelligence’s database migrations inside its own crate so the published
databasefeature builds without a sibling HTTP checkout. Preserve all historical migration filenames and SQL bytes for existing databases. (#632) -
[SDK] Java, Python, Go, and browser dynamic-mock clients use the HTTP serving port and
/__mockforge/api/mocks. Java sends the server’s mock schema and discovers assigned ports; Python ignores port-zero announcements and cleans up failed starts. Failed management requests surface errors instead of appearing to create local-only stubs. Preserve management-token support on the corrected requests. Browser package builds now gate on strict type checking, and runtime CI exercises real served responses and cleanup. ForgeConnect initializes without requesting environments when no workspace is configured; explicit workspace failures remain visible. Browser tests use Vitest with an audited dependency tree and a separate live-server suite. (#1281, #1282, #1011) -
[AI] AI Studio mock and persona
GenerateOnceFreezecalls now find and reuse snapshots by their recorded input identity, honor the configured snapshot directory, and verify payload integrity. First use no longer fails when the directory is missing. Snapshot identity includes workspace, model and effective seed, and concurrent publication preserves one completed snapshot. Live-provider sampling remains best effort. (#852) -
[UI] Plugin, scenario and template publication encode file bytes in bounded chunks instead of spreading the entire file into a function call. Large uploads no longer fail with a JavaScript argument-stack
RangeErrorbefore reaching the API. -
[Cloud] Cloud test runs no longer get stuck in
runningwhen the registry rate-limits the runner. The runner’s start, event and finish callbacks all come from one IP, so the per-client limit could return 429 to/finishand the run never ended. Internal callback routes now skip the rate limit when they carry the internal API token (requests without it are still limited), and the runner retries/finishon 429, 5xx and connection errors with backoff (6 attempts over about 30 s). (#1286) -
[DevX] Removed
crates/mockforge-cli/src/registry_commands.rs. No module declared it, so it was never compiled (it would not have compiled), and itsplugin registrycommands never existed.docs/PLUGIN_MARKETPLACE_IMPLEMENTATION.mdnow lists the real commands. (#1277) -
[Cloud] The SQLite registry store (self-hosted registry admin) now rotates API tokens and lists tokens due for rotation, like the Postgres store. Before, rotation returned an error and the rotation reminder list was always empty. Both stores now generate
mfx_tokens in one URL-safe format (32 random bytes, unpadded URL-safe base64), so new Postgres tokens no longer contain+,/or=. Existing tokens keep working. (#1266) -
[Docs] The README and
cloud-service/README.mdno longer claim a multi-region deployment or a 99.9% uptime SLA. Production runs in one region (US East) with no contractual SLA. The README now says which RAG storage backends persist: file storage does; the database backend falls back to memory, and vector databases are not built in. It also drops the line saying every documented feature was verified. (#1265)
Security
- [DevX] Raise Wasmtime and WASI to 36.0.20, with the matching Cranelift, Pulley and Wiggle family, to address RUSTSEC-2026-0424 (unchecked memory usage when transferring borrowed resources between components). Both plugin crate manifests require the patched minimum.
- [DevX] Clear the 2026-10-29 RustSec advisory batch: patch-bump
pulldown-cmarkto 0.13.5,libz-systo 1.1.31 andevent-listenerto 5.4.2 (RUSTSEC-2026-0414, -0418, -0221), and add documentedaudit.tomlignores for the rest — the zstd-safe 8.x-fixed advisories on the wasmtime-36.x-pinned 7.2.4 copy, transitive vulnerabilities whose only fix is a major bump the dependent doesn’t allow yet (glib, lru, git2, the rand 0.7 copy), and unmaintained-crate notices with no upstream fix. - [DevX] Bump
zstdto 0.14.3 andzstd-safeto 8.2.0 (andzstd-systo 2.1.2) to patch RUSTSEC-2026-0391 (stream-reuse data leak), -0398 and -0403 (zstd-safe OOBVeclength and uninitialized memory) on the request/proxy path (async-compressionin mockforge-proxy/http). The only copies left below the fix arezstd0.13.3 /zstd-safe7.2.4, pinned by wasmtime 36.x’s internal cache (not the request path) and by mockforge-kafka, which have no in-line patch; those three advisories are ignored for that copy inaudit.tomluntil the wasmtime major bump, with a mockforge-kafka zstd 0.14 bump tracked as a follow-up. - [DevX] Raise Wasmtime and WASI to 36.0.19, including the patched Cranelift family, to address RUSTSEC-2026-0379 and RUSTSEC-2026-0380. Enforce the patched minimum in both plugin crate manifests. Update rustls-webpki to 0.103.16 and webpki-roots and webpki-root-certs to 1.0.10 (RUSTSEC-2026-0382, -0381, -0383). rustls-webpki 0.102.8 has no patched release and is still pulled in by rumqttc 0.25.1 and async-nats 0.43, so RUSTSEC-2026-0382 is ignored for that copy in
audit.tomluntil they upgrade. - [DevX]
{{encrypt}},{{secure}}and{{decrypt}}no longer fall back to a password written in the source (mockforge_default_encryption_key_2024). With no key-store key and noMOCKFORGE_ENCRYPTION_KEY, they output<encryption_error: no key configured>or<decryption_error: no key configured>. The key derived fromMOCKFORGE_ENCRYPTION_KEYnow uses a fixed salt, so{{decrypt}}can read what{{encrypt}}produced. Before, PBKDF2 used a new random salt on every call and decryption always failed. The uncompiledtoken_resolvers.rs,plugin_integration.rsandintegration_demo.rsare removed from mockforge-core. (#1273) - [DevX]
mockforge-k8s-operatormoves to kube 4.2, k8s-openapi 0.28 and schemars 1. kube 0.87 was the last dependency on rustls 0.21, which has no patched release for RUSTSEC-2026-0346 (Debug output exposes session secrets) or RUSTSEC-2026-0349 (ServerHello downgrade). The workspace no longer builds rustls 0.21, and both advisories are no longer ignored inaudit.toml. The CRD manifests ink8s/crd/are unchanged. (#1232) - [Cloud] Upgrade
zipto 9.0.4 in mockforge-cli, plugin-loader, plugin-cli, scenarios and registry-server (RUSTSEC-2026-0367/0368/0369). MockForge’s own archive handling already avoidedextract()and checkedenclosed_name(); entry names that cannot be decoded are now skipped. The desktop app’s auto-updater still bundles zip 4.6.1, which only unpacks signature-verified MockForge updates. - [Cloud] The registry now enforces
mfx_API token scopes on every token-authenticated route. Before, only plugin publishing checked them, so any valid token could deploy, update or delete hosted mocks, run tests and use other cloud resources. Hosted mocks and other cloud mock resources needdeploy:mocks. Token management, account security, SSO, GDPR, platform admin and promotion approvals reject tokens. Routes without a declared scope reject tokens. Tokens are bound to their own organization: an{org_id}path orX-Organization-Idnaming another org returns 403, and so does anyX-Organization-Slug. Breaking for tokens that relied on the missing check. Tokens created withoutdeploy:mocks(the column default isread:packages) now get403 INSUFFICIENT_SCOPEnaming the missing scope. Re-issue them with the needed scope. (#1263) - [Cloud] Registry migration
20250101000102disables the seed accounts that20250101000002created with published passwords:[email protected]/admin123(an admin) and[email protected]/test123(#1014; the earlier fix was closed but never reached main). Only rows that still carry a seed password hash are touched. Their password stops working, admin is cleared, refresh tokens are revoked, and the legacy API token and OAuth links are removed. Rows are disabled rather than deleted because organizations reference them withON DELETE RESTRICT. It applies at the next registry boot.make seedis replaced bymake dev-seed, which requiresMOCKFORGE_DEV_SEED=1and a loopback database and prints a random password once. - [Cloud] The registry accepts
?token=as a credential only on the three SSE stream routes (hosted-mock runtime logs, test-run events, test-generation jobs), which browsers’EventSourcecannot send headers to. Before, every authenticated route accepted it, so tokens could be put in ordinary URLs and end up in proxy logs, browser history and referrers. The check uses the route template axum matched, so trailing slashes, extra segments or encoded variants of a stream URL don’t widen it, andmfx_API tokens are never accepted from a URL. Request logs and trace spans now redacttokenandaccess_tokenquery values on every route. Header and session-cookie auth are unchanged. Breaking for a client that put a token in the query string of a non-stream route; sendAuthorization: Bearerinstead. No client in this repo did. (#1018) - [DevX] Admin server reads need the same credentials as writes. Request logs, recorder state, config and every other
GET /__mockforge/*route answered anonymous requests (the #1019 fix was closed but never reached main). Bound to anything but loopback, the admin server now refuses to start withoutMOCKFORGE_ADMIN_PASSWORD,MOCKFORGE_MANAGEMENT_TOKENorJWT_SECRET. On loopback with nothing configured, anonymous reads still work from the same machine (loopbackHostrequired). The overrides API and/api/resilience/*are no longer exempt. The JWT fallback secret, previously a constant in the source, is random per process. Log and config responses redact credential headers and URL userinfo. The registry sends each hosted mock’s management token on its private-network calls. (#1019) - [DevX] The admin UI never creates built-in users.
admin/admin123,editor/editor123andviewer/viewer123were seeded wheneverJWT_SECRETwas unset outside development, and in dev-auth mode (#1016; the earlier fix was closed but never reached main). The only login is now an operator-setMOCKFORGE_ADMIN_PASSWORD; without credentials a loopback admin server allows anonymous reads and any other bind refuses to start. The dev-auth JWT secret no longer falls back to a constant in the source. The login page no longer offers demo accounts. - [DevX]
{{fs.readFile}}only reads files under one root:MOCKFORGE_TEMPLATE_FS_ROOT, or by default the directory of the spec given tomockforge serve. It refuses.., percent-encoded and Windows-style paths, symlinks that escape the root,/proc,/sys,/dev,/etcand env-like files, and files overMOCKFORGE_TEMPLATE_FS_MAX_BYTES(default 1 MiB). It is disabled when no root is configured, for specs fromMOCKFORGE_CONFIG/MOCKFORGE_OPENAPI_SPEC_URL, and on hosted mocks unlessMOCKFORGE_TEMPLATE_FS_ROOTis set. Pinned directory handles reject symlink replacements after path checks, and nonblocking opens reject named pipes without waiting for a writer. Previously any spec could read any local file into a response. (#1015) - [DevX]
mockforge plugin installandplugin updatenow refuse a plugin whose signature is missing or fails verification. Before, a failed check only logged a warning and the plugin installed anyway, and--skip-validationskipped the check entirely. Signatures (format v2) now cover every file in the plugin directory, not onlyplugin.toml, so a swapped WASM module, an added file or aplugin.sigcopied from another plugin fails. Keys can be revoked or given an expiry in a trust store (--trusted-keys FILE,$MOCKFORGE_PLUGIN_TRUSTED_KEYS_FILEor~/.mockforge/trusted-keys.json).--allow-unsignedinstalls a plugin with noplugin.sig, for local development only, with a warning; a signature that is present but invalid is still rejected. Breaking: unsigned plugins need--allow-unsigned, v1 signatures must be re-signed,--no-verifyis now an alias of--allow-unsigned, andInstallOptions::verify_signatureis replaced byallow_unsigned. The admin installer now enforces the same trust, revocation, expiry and signature policy; its unsigned-development checkbox starts unchecked. Both installers load the exact verified directory rather than rediscovering another plugin with the same ID. WASM data checks now parse the original binary instead of Wasmtime’s serialized native code, which rejected valid plugins. (#1013, #1276) - [DevX] CORS fails closed. With no configured origins (now the default), the HTTP mock server grants no cross-origin access. Before, it fell back to
Access-Control-Allow-Origin: *when origins were unset, when any entry failed to parse, or when more than one was listed. Several exact origins now each work, a malformed entry denies all,*is never combined with credentials, andlocalhostenables an explicit dev mode. NewMOCKFORGE_CORS_ALLOWED_ORIGINSsets the list. The admin UI no longer allowsOrigin: nullby default and refuses*inCORS_ALLOWED_ORIGINS. Hosted mocks and the demo set*explicitly. Upgrade note: browser apps calling a self-hosted mock from another origin needMOCKFORGE_CORS_ALLOWED_ORIGINS(for examplelocalhost) orhttp.cors.allowed_origins. (#1017) - [Cloud][DevX] MockForge’s control plane on the HTTP port now needs a management credential for reads as well as writes, on hosted and self-hosted servers. This covers
/__mockforge/*(management API, chains, contract diff, fixtures, MockAI, time travel, route chaos, network profiles, scenarios, the live stream),/api/chaos,/api/recorder,/api/verification,/api/world-state,/api/performanceand the/api/v1/*management APIs. Before this, self-hosted servers served all of it to anyone who could reach the port (it always bound0.0.0.0), and hosted mocks served every read anonymously. SetMOCKFORGE_MANAGEMENT_TOKEN(32+ characters) and send it asX-MockForge-Management-Token; the admin user’s Basic credentials also work. Without a credential the control plane answers 401.MOCKFORGE_INSECURE_MANAGEMENT=1turns the check off only whenMOCKFORGE_HTTP_HOSTis a loopback address; otherwise the server refuses to start. Your mocked API,/health*and/__mockforge/docsstay public. The HTTP server now bindsMOCKFORGE_HTTP_HOSTinstead of always0.0.0.0. The CLI, SDKs, IDE and browser extensions and the cloud UI send the token. A mock that shares a control router’s prefix (for examplePOST /api/v1/auth/login) is no longer caught by the hosted-mock write gate. (#1011)
[0.3.236] - 2026-10-16
Added
- [DevX]
bench-chunked --request-timeout SECSbounds each request end to end (connect, chunked upload, response). The default is twice the minimum upload time plus 60 s, and0disables it. Timed-out attempts count as failed and in a newtimed_outfield, and the console showsFailed: N (M timed out after Xs). Previously no timeout was set, so a silently dropped upload held its worker for minutes until OS TCP timeouts fired. (#79) - [DevX]
bench-chunked --spec ... --overlap-operationsstarts each operation when the previous one’s--durationwindow closes instead of after its in-flight uploads finish, so request starts stay continuous at--rpsacross operations in long campaigns. With it,elapsed_secondsis the pass’s wall clock. (#79)
Changed
- [Contracts] Move drift-budget evaluation into Contracts behind typed fitness and consumer-impact providers. Preserve Core constructors, concrete registry setters and evaluation entrypoints, while allowing cross-crate consumers to use the Contracts-owned engine. (#636)
Fixed
- [Cloud] Registry S3 storage with a custom endpoint (MinIO, Cloudflare R2) uses path-style addressing, so MinIO buckets verify instead of silently falling back to local storage. New
STORAGE_REQUIRE_S3=truefails startup rather than fall back. (#1236) - [Cloud] Registry artifact downloads from the private bucket use presigned URLs. Spec URLs handed to the test runner are presigned only for the caller’s own uploads. Publishing can no longer overwrite another owner’s plugin, scenario or template, including under RLS, and published versions are immutable (409). (#1239)
Security
- [Cloud] Remove most of the rustls 0.21 dependency graph (RUSTSEC-2026-0346/0349): the AWS SDK S3/KMS clients drop their legacy hyper 0.14 TLS stack,
oauth2no longer pulls reqwest 0.11 (registry OAuth code exchange uses the workspace reqwest client with redirects disabled), the SMTP STARTTLS listener moves to tokio-rustls 0.26 with an explicit ring provider, and unused TLS dependencies are dropped from mockforge-tcp. The Kubernetes operator’s kube 0.87 client is the only remaining rustls 0.21 user. (#1232)
[0.3.235] - 2026-10-15
Security
- [DevX] Update rustls to 0.23.46 (RUSTSEC-2026-0346/0347/0348/0349) and wasmtime/wasmtime-wasi to 36.0.18 (RUSTSEC-2026-0352/0363). The remaining rustls 0.21 copy has no patched release; its removal is tracked in #1232.
Fixed
- [DevX]
bench-chunkednow reports the request-start rate alongside the overall rate, so short runs can be checked against--rps. Summaries addstart_window_rps(attempts divided byload_window_seconds, the sum of each operation’s--duration) andupload_mbps; exported operations addstart_rate_per_sec,load_window_msandupload_mbps. The console replaces the ambiguousThroughput: N req/sline withRPS (started),RPS (overall)(including the in-flight drain) andUpload: N Mbps. Existing fields are unchanged. (#79)
[0.3.234] - 2026-10-14
Added
-
[DevX] Export live local workspace configuration to versioned JSON/YAML with credential references and stable ordering/timestamps. Authorized imports create new workspaces without overwriting existing state; explicit activation publishes namespaced static HTTP mocks atomically while retaining unrelated mocks. Unsupported authentication activation is rejected. (#1197)
-
[Cloud] Hosted deployments can explicitly select or change their capture workspace in the API and portal. Recorder batches use the deployment’s verified organization/workspace, preserve historical attribution, and reject cross-organization assignments; unassigned deployments stay unassigned. (#720)
-
[Contracts] Add the
proxyCLI command with passive or strict OpenAPI conformance validation and the shared violation feed. Preserve binary and redirect responses, honor configured upstream timeouts, and validate response status ranges, default schemas, and malformed JSON. (#864) -
[Cloud] Saved observability queries can raise tenant-scoped incidents from request logs, traces, and recorded exchange metrics. Validated alert rules share centralized severity mapping and IncidentBus dispatch; repeated breaches deduplicate, sampled recovery resolves, and missing or incomplete telemetry does not imply recovery. (#720)
-
[Cloud] Monitored services support per-service cron schedules and IANA timezones, with legacy interval cadence preserved. Probe runs and due-state claims commit atomically so overlapping scheduler ticks do not enqueue the same occurrence twice. (#720)
Changed
-
[DevX] Failure-designer and incident-replay HTTP handlers now live in Intelligence behind typed Chaos providers, preserving the existing HTTP states, request types, generated scenarios, import formats, and error responses. (#632)
-
[DevX] Snapshot diff routes and comparison logic now live in Intelligence behind a typed current-mock projection. HTTP retains its management state, public DTO paths and router signature. (#632)
-
[DevX] Risk-simulation HTTP handlers now live in intelligence behind a typed engine interface. Shared risk-assessment DTOs live in foundation, while HTTP retains its original state fields and public auth type paths. (#632)
-
[DevX] Behavioral-cloning HTTP handlers now live in intelligence and use shared foundation scenario types. A recorder storage adapter preserves the HTTP compatibility state and SQLite behavior without adding a reverse dependency. (#631)
Fixed
-
[DevX] Return completed UUID template expansions before running unrelated token scans. Keep fresh UUID generation, repeated-token behavior, unresolved tokens, and bare response-function processing.
-
[DevX] Keep pre-commit Clippy in one invocation with the complete changed-file list. This avoids overlapping target-lock queues and repeated feature graphs while preserving crate/workspace selection and lint flags. (#706)
-
[Cloud] Enforce one case-insensitive email identity across concurrent registry signups and account lookups. Reject legacy collisions without merging accounts; clear verification after an email change. OAuth requires provider-verified email proof and rejects unverified local-account linking or replacement of an existing provider identity. (#908)
-
[Protocols] AMQP pull deliveries and committed transaction publishes are recorded; rolled-back publishes and empty pulls remain excluded. MQTT captures retained deliveries and last-will publishes and deliveries. Publisher confirmations use a separate sequence from consumer delivery tags, allowing publish and consume operations on the same channel. Real AMQP, MQTT and Kafka clients verify durable recording payloads and metadata. (#715)
-
[DevX] Honor optional CLI protocol and subsystem features during shared startup, preserve standalone listeners and shutdown handling, and reject unavailable requested operations before creating runtime state. Declare HTTP and VBR prerequisites and include SMTP in its HTTP query-import gate.
-
[DevX] gRPC startup accepts an owned listener and retains its socket through discovery and TLS setup. Real client fixtures reserve loopback port-zero listeners instead of releasing a temporary port before server startup.
-
[DevX] OIDC environment fixtures run in isolated child processes, preventing parallel tests from overwriting issuer and discovery settings while exercising the real environment-based configuration paths.
-
[Contracts] Move drift-budget and contract-health handlers into Intelligence through typed engine providers, retaining the existing HTTP state constructors and feature gates. Bind timeline workspace IDs as UUIDs so persisted threat and forecast rows appear instead of an empty successful response. (#632)
-
[DevX] Preserve the public HTTP handler module paths for consistency, fidelity and X-Ray after moving their implementations to Intelligence. (#632)
-
[DevX] Honor optional HTTP scenario, world-state, performance and route-chaos feature gates for modules, state and route wiring, allowing minimal builds to retain basic mocks and health probes.
-
[DevX] Preserve shared configuration types in minimal Core builds and gate optional workspace and advanced runtime integrations consistently. Voice workspace construction declares its workspace/data prerequisites, and advanced engines declare their workspace and persona-type prerequisites.
-
[DevX] Integration builds check shared-disk headroom and restored Cargo library artifacts before compiling. Read-only diagnostics replace automatic target deletion and retain foreign or active build files. (#706)
-
[DevX] Integration preflight verifies an owned 1 MiB temporary allocation and flush, reporting per-user quota and space errors before build retries. Existing temporary files and caches are preserved. (#706)
-
[DevX] Publish Recorder before the AMQP, Kafka and MQTT crates that depend on it. The publish-list guard checks normal and build dependency order, preventing partial releases caused by missing same-version dependencies.
-
[VBR] Create SQLite foreign keys with table declarations and enforce them on every connection. Return created records using their actual keys with HTTP 201. Runtime tests use isolated databases and fail on setup errors instead of silently passing.
-
[Security] Update the locked QUIC protocol dependency to
quinn-proto0.11.18, fixing five RustSec packet-handling and memory-exhaustion advisories. Its required Rand 0.10 patch dependency is updated without enabling new protocol features. -
[Cloud] Contract-drift AI callbacks carry the persisted run identity so usage audits identify the originating user. The registry verifies the run, suite, and sampled workspace belong to the requested organization and rejects forged actor claims; scheduled and older anonymous callbacks remain anonymous. (#869)
-
[Cloud] Hosted-mock spec upload uses a keyboard-accessible native button, and deployment table headers expose column semantics. Cloud configuration and dashboard browser tests follow their supported workspace features, while local server controls retain separate coverage. (#1051)
[0.3.233] - 2026-10-12
Fixed
-
[DevX] Explicit mock health routes coexist with built-in health probes instead of crashing server startup; this restores the generated
initproject’s OpenAPI example. (#851) -
[DevX] The README opens with a working Docker example and a reproducible cold-start measurement. New
initprojects print their startup command and test request; installation guidance no longer promises a 30-second source build. (#851) -
[Cloud] HTTP hook response previews truncate at a valid UTF-8 boundary, preventing logging from panicking on multibyte response bodies in both orchestration implementations.
-
[DevX]
ftp serve --configreads standalone FTP settings or a server YAMLftpsection, loads configuration-relative fixture directories, and honors explicit CLI overrides. FTP listeners apply the configured passive port range and idle timeout. (#794) -
[Cloud] Stripe trials allow checkout without a payment method and cancel when the trial ends without one; zero-trial paid checkout still collects a card. Actual SDK request tests cover both paid plans and billing intervals. (#837, #714)
-
[Reality] Verify persona CI-hit recording from actual HTTP headers through SQLite persistence, including optional run IDs and ignored blank/invalid labels, and document analytics setup and tenant-context limits. (#716)
-
[DevX] Retire the outdated shared-runner cleanup copy and its systemd units in favor of the infrastructure-owned script, preventing accidental replacement of newer host protections. (#706)
-
[Contracts] Swagger 2.0 imports use automatic conversion instead of rejecting the format. Reusable parameters/responses, inherited request bodies, nested schema references, and media-specific response examples survive loading and mock generation. (#838)
-
[DevX] Intelligence handler state construction tolerates Cargo database feature unification, restoring HTTP builds when Core enables database support independently. Explicit HTTP database connections are preserved. (#632)
-
[DevX] Documentation checks run on hosted runners and require only enabled mdBook preprocessors, freeing the dedicated Rust runners for tests and publication.
-
[DevX] Chaos ML and orchestration crates document their public API and enforce missing-documentation warnings through workspace lints, including explicit limits of simulated cluster execution and local manifest synchronization. (#799)
-
[DevX] Crate publication returns success after every package is available and fails when any package remains unpublished or its registry lookup fails.
-
[Reality] HTTP pipelining benchmarks stop at the configured load deadline, retain partial wire-byte counts, and drain large responses while sending subsequent requests. Response framing handles interim replies, chunked bodies, HEAD, and connection-close bodies. Multipart synthetic files accept
--multipart-content-type. (#937)
[0.3.232] - 2026-10-10
Fixed
-
[Contracts] Successful JSON Schema and OpenAPI 3.1 validation uses the compiled fast path while invalid requests retain every validation diagnostic. OpenAPI extension checks remain enforced. Simple validation benchmarks improved by about 75% locally and the CI regression gate passes. (#1165)
-
[Cloud] Flow row actions use responsive confirmations and bounded live event streams; State Machines history no longer traps sidebar navigation in a render loop. Terminal test runs have authorized deletion in the API and portal. (#1164)
-
[Cloud] Password recovery screens, visible failed-login and invalid-email feedback, correct sign-out navigation, and page refresh and heading cleanup improve portal account workflows. (#1164)
-
[Cloud] PostgreSQL password reset requests bind the expiry parameter correctly, restoring one-hour recovery links for existing accounts.
-
[DevX] Crate publication identifies its crates.io API requests so index checks avoid rejected requests and unnecessary per-crate waits.
-
[DevX] Pull-request benchmarks compare the immutable base and candidate in adjacent, order-balanced measurements on an isolated runner. The 5% regression limit stays enforced without relying on a moving post-merge baseline.
[0.3.231] - 2026-10-09
Added
- [Reality]
bench-chunked --raw-bodysends the original raw filler while preserving any declared Content-Type, for negative WAF testing. Chunked reports distinguish per-round elapsed time, configured rate limits, and measured per-target request throughput. (#79)
Changed
- [DevX] Breaking build requirement: The minimum supported Rust version is now 1.96.0, matching the pinned build toolchain. Workspace packages declare this minimum and CI verifies the actual compiler and Cargo versions. Users building from source or using
cargo installon older Rust versions must upgrade first. Prebuilt binaries and container images do not require a local Rust toolchain.
[0.3.230] - 2026-10-04
Changed
- [Cloud] Audit logs no longer record email addresses, usernames or SAML NameIDs. Member, invitation, SSO login and password-reset events identify people by user id or invitation id; failed logins store a keyed hash of the attempted email instead of the address. (#1087)
- [Cloud] Audit logs are now kept for 400 days and then deleted by a daily retention worker. The append-only trigger allows only this purge, and only for rows past the window;
verify_chaintreats the oldest surviving row as the chain start. (#1087) - [Cloud] Privacy Policy 2.2 and DPA 2.2 describe audit-log retention after account deletion. Terms 2.1 name Delaware as governing law and venue.
- [DevX] The Kubernetes manifests (
k8s/) and the Helm chart now give every MockForge replica its own recorder volume. Previously all replicas mounted one ReadWriteOnce claim holding the recorder’s SQLite database, so pods on other nodes could not start and pods on the same node shared one SQLite file.k8s/deployment.yaml+k8s/pvc.yamlare replaced byk8s/statefulset.yaml(a StatefulSet with avolumeClaimTemplatesentry, onerecorder-db-mockforge-Nclaim per pod), and the HPA now targets it. The Helm chart renders a StatefulSet (plus a<fullname>-headlessService) whenpersistence.enabledis true, the default, and a Deployment without a recorder volume when it is false. Upgrading an existing Helm release replaces its Deployment with a StatefulSet and creates per-pod claims (data-<fullname>-N); the old shared<fullname>-dataclaim is kept (annotatedhelm.sh/resource-policy: keep) but no longer mounted, so copy any recordings you need out of it, then delete it. The Kubernetes test workflow’s Chaos Engineering job, which never ran, now runs on manual dispatch and weekly.
Fixed
-
[Reality]
bench-chunkedsends a valid JSON body when the request’s Content-Type is JSON (application/jsonor any+jsontype, from--headeror a targets-fileheadersentry). The body is the spec-generated request body plus a_paddingstring field, sized to exactly--total-size-bytesand streamed in the same chunks. Previously every body wasXfiller, so WAFs flagged each request as malformed JSON. Other content types still get filler. (#79) -
[DevX] Registry audit email hashing compiles under strict unused-qualification lints in both PostgreSQL and SQLite builds. Hashing behavior is unchanged.
-
[Cloud] Flow creation sends the registry-required configuration across all five editors. Federation details have a labeled Back button; portal dialogs, placeholders and disabled controls use readable theme colors and wrapping action rows, and service route counters retain their spacing.
-
[Cloud] Authenticated portal workflows recover from deleted workspace selections and page errors; resilience and snapshot comparisons load, deletion dialogs work, fixture lists refresh after writes, and fitness authoring sends supported evaluator settings. Cookie reloads restore tokens before loading protected pages. Theme toggles, deployment slugs, actor labels and action accessibility are corrected. (#1144)
-
[DevX]
cargo install --locked mockforge-clino longer warns about yanked dependencies (chacha200.10.0 andspin0.9.8/0.10.0 bumped to their patched releases). -
[DevX] GDPR erase on the SQLite backend did nothing (
delete_user_data_cascadereturned 0). It now matches the Postgres store. (#1087)
[0.3.229] - 2026-10-01
Changed
- [Cloud] Tenant isolation is now also enforced in the database for every org-scoped table (36 tables under Postgres row-level security, up from 5). Request-path queries run bound to the caller’s org; background workers, webhooks and platform-admin paths run on the owner role. Requires the owner role (
DATABASE_URL) to haveBYPASSRLS; the migration refuses to apply otherwise. Startup now logs which roles back each pool and flags a misconfigured one at ERROR. (#1087) - [Cloud]
PUT /api/v1/organizations/{org_id}/quotais now platform-admin only. Org owners can no longer set their own quota overrides, because overrides are merged over plan limits.GETnow requires membership in the org. (#1087) - [Cloud] Row-level security now also covers 38 tenant tables that have no
org_idcolumn of their own (workspace, hosted-mock and other child tables). Each is isolated through its parent’s org, and per-user tables are scoped with a newapp.current_user_idsetting. (#1087, #1120) - [Cloud] Requests without an organization header now resolve to an org the user belongs to even when they own none (SSO-provisioned members, or users who deleted their personal org). Previously every such request returned 404. (#1087, #1117)
- [DevX] Kubernetes manifests and the Helm chart now meet the “restricted” Pod Security Standard (non-root uid 999, no privilege escalation, all capabilities dropped), and liveness/readiness probes use the HTTP port instead of the auth-protected admin port. (#1118)
- [DevX] Dependency refresh: 33 patch and minor updates, including tokio 1.53.1 and hyper 1.11. (#1104)
Fixed
- [Cloud] The registry API no longer occasionally drops a request mid-flight (clients saw a closed connection or 502). The request-logging middleware held a tracing span open across an
await, which could panic once a request moved between worker threads. - [Cloud] Security: any authenticated user could read or overwrite any org’s quota overrides. (#1087)
- [Cloud][AI] Security:
POST /api/v1/organizations/{org_id}/mockai/generate-openapi-from-trafficdid not check that the caller belongs to the org. (#1087) - [Cloud] Security: the GDPR data export included pending invitation payloads, which let a plain member redeem an invite meant for someone else. (#1087)
- [Cloud]
GET /api/v1/organizations/{org_id}/incidents/statsno longer returns 500 once the org has a resolved incident. (#1087) - [Cloud] Security: a workspace could aim a chaos campaign at another org’s hosted mock and inject faults into it. Campaign targets are now checked against the caller’s org at create, trigger and snapshot restore, and the internal chaos toggle requires the run that owns it. (#1124)
- [Cloud] Security: adding a capture to a capture session did not check who owned the capture, so a replay could read another org’s captured request bodies. (#1124)
- [Cloud] Capture sessions, clone models, monitored services, diff runs and fitness functions belonging to another org now return exactly the same “not found” response as missing ones. (#1087, #1128)
- [Reality][AI] MockAI no longer stores a session for every write request that arrives without a session ID. Those sessions were never read again and every write waited on one shared lock, so throughput fell steadily under sustained load (about 2x faster writes, flat memory). Per-request fixture-matching logs are now at debug level instead of warning. (#1125)
- [Cloud] Returning visitors no longer see “Something went wrong” after a cloud UI deploy: the app’s entry JS and CSS are content-hashed so browsers can’t keep a stale copy. (#1116)
[0.3.228] - 2026-09-30
Added
- [Reality] The mock Kafka broker can filter messages so it counts or verifies them without keeping every payload in memory. (#992)
Fixed
- [Cloud] Hosted mocks now require a per-deployment management token for writes to MockForge’s control routes (
/__mockforge/*,/api/chaos,/api/recorder, and the other management APIs). Previously anyone who knew a hosted mock’s URL could change it. Reads and your mocked API are unaffected, and self-hosted servers don’t use a token. Owners can reveal the token on the Hosted Mocks page. (#1105) - [DevX] Windows release binaries build again (a Unix-only socket option in
bench-qosbroke every Windows build since 0.3.223). (#1101)
[0.3.227] - 2026-09-30
Added
- [Reality]
mockforge bench-chunkedgains--rps N(per-target cap on request starts per second),--cps(new TCP/TLS connection per request), and campaign mode:--rounds N/--repeat-until <duration>re-run the whole pass into<output>/round_N/, with per-round stats incampaign.jsonlandround-summaries/, and--keep-rounds Nrotating old round dirs, matchingmockforge bench. (#79) - [Cloud] Redesigned app shell for the logged-in portal: new sidebar, top bar, search and menus, refreshed shared UI primitives, and a new dashboard; every page moved onto the new frame. (#1086)
Fixed
- [Cloud] GDPR account erasure no longer fails for users who have audit-log rows. (#1093)
- [Reality]
bench-chunked --chunk-interval-msnow sends the first chunk immediately and waits only between chunks. Previously it waited before the first chunk and sent the last two back to back. (#79)
[0.3.226] - 2026-09-29
Added
- [DevX] Response overrides now work in
mockforge serve. Rules come from a newoverrides:list inmockforge.yaml,MOCKFORGE_HTTP_OVERRIDES_GLOBfiles, and a newMOCKFORGE_HTTP_OVERRIDESJSON array. With--admin,GET/PUT /__mockforge/overridesreads and replaces the live rules without a restart; an invalid set returns 400 and keeps the previous rules. Rules gainnameandenabled. (#1088) - [DevX] Admin UI Overrides page to toggle, reorder, edit, and save override rules for the running mock (self-hosted) or a hosted mock (cloud). (#1095)
- [Cloud] Per-hosted-mock override rules:
GET/PUT /api/v1/hosted-mocks/{id}/overridesstores rules on the deployment, passes them to new machines, and pushes them live to running ones over the private network. (#1094)
Fixed
- [DevX]
tag:override targets match the operation’s real OpenAPI tags, andwhen:conditions see request headers, query, and body. Invalid JSON pointers are rejected instead of silently patching the whole body, and regex targets work in rule sets loaded from config (including gRPC). (#1088) - [Cloud] Redeployed hosted mocks now start their admin server, so the resilience dashboard and override push can reach them. (#1094)
- [Cloud] Plugin registry search no longer returns 500 for the default “popular” sort and the “security” sort. (#1082)
- [DevX] The admin UI top-bar search suggests pages and navigates on Enter. (#1082)
- [Cloud] Bump wasmtime and wasmtime-wasi to 36.0.16 for RUSTSEC-2026-0316 and RUSTSEC-2026-0314. (#1092)
[0.3.225] - 2026-09-27
Added
- [Reality]
mockforge bench-chunked --targets-file <file>sends realTransfer-Encoding: chunkedtraffic to every target in a multi-target file in parallel (--max-concurrency, default 10). It uses the same file format asbench --targets-file, including per-targetauth/headers/spec. Per-target artifacts go to<output>/target_N/, with a roll-up inchunked-multi-target-summary.json. (#79)
[0.3.224] - 2026-09-24
Added
- [Reality] Campaign mode now supports
--no-k6-logs(skipsk6-output.logand conformance debug sidecars; k6 output is drained instead of buffered) plus per-round aggregate artifacts:round_N/round_summary.json,round_N/all_targets.csv, durablecampaign.jsonl, andround-summaries/copies. New--keep-rounds Nprunes old round directories during a campaign while preserving campaign-level stats, and k6 scripts are generated once per target and reused across rounds. (#79)
[0.3.223] - 2026-09-18
Added
- [Reality] Multi-target
--repeat-until <duration>and--rounds Ncycle the full targets list so low concurrency can still stress every server over a longevity window without manual restarts. Pair with a short--duration(per-batch k6 lifetime). (#79)
[0.3.222] - 2026-09-16
Fixed
- [Reality] Longevity / huge-spec k6 runs no longer emit a Trend+Rate pair per OpenAPI operation by default when ops >= 500 or duration >= 1h (
--per-op-metrics/--no-per-op-metricsto force).--max-concurrencyauto-caps to 3 for huge specs, and SIGKILL gets an OOM hint. (#79)
[0.3.221] - 2026-09-12
Fixed
- [Contracts] OpenAPI 3.1 schema
typearrays (["string", "null"]) are coerced to OAS 3.0type+nullableso the spec loads. LLM-generated specs failed withinvalid type: sequence, expected a stringand never extracted operations. (#79)
[0.3.220] - 2026-09-12
Fixed
- [Reality]
traffic-breakdown.jsonno longer repeatsunique/total/expected_requestsnext tounique_cases/projected_*. Those aliases were a one-release bridge and got read as k6 wire counts.unique_casesis still the YAML case count (not traffic on the wire).projected_per_second/projected_over_runstay the plan. (#79)
[0.3.219] - 2026-09-10
Fixed
-
[Reality] HTTPS + ALPN HTTP/2 was rejecting WAF
Connectionheaders (http2: invalid Connection request header). The header stays on the wire (it is the hop-by-hop test); k6 is forced onto HTTP/1.1 viaGODEBUG=http2client=0when--wafbench-verbatimis set or any request has aConnectionheader. generate-only prints the same prefix for a manualk6 run. (#79) -
[Reality]
traffic-breakdown.jsonfields are the plan, not k6 counters:unique_cases(YAML case count),projected_per_second(unique_cases * rps),projected_over_run(unique_cases * rps * duration).unique/total/expected_requestsstay as aliases for one release.expected_requests_unitis dropped. (#79)
Added
- [Reality] Multi-target runs print estimated wall clock:
ceil(targets / max-concurrency) * duration.--vusand--rpsare per target, not shared. (#79)
[0.3.218] - 2026-09-07
Fixed
-
[Reality] Static k6 paths are JSON-encoded and concatenated onto
BASE_URL, so WAF URIs containing\x(Werkzeug UNC/static/\\attacker.com\share\x) no longer land in a JS template literal. k6/goja treated\xas a hex escape and exited before sending any request (invalid escape: \x: len("") != 2). The wire URI is unchanged.validate_scriptnow rejects a bare\xhex escape in-process. (#79) -
[Reality]
traffic-breakdown.jsonfieldexpected_over_durationis nowexpected_requests(HTTP count over the run, not seconds). When--rpsis unset the value isnullinstead of inventing rps=1, which made a 60s run look like a duration of 300. (#79)
[0.3.217] - 2026-09-04
Fixed
-
[Reality] OpenAPI header parameters named
Content-Length,Host, orTransfer-Encodingare no longer invented from the schema (#79). Srikanth’s Amazon S3 spec listedContent-Lengthas an integer header; generation filled42on empty-body POSTs and k6 dropped every request (Content-Length header \"42\" doesn't match actual body length of 0). k6 now owns those headers. Verbatim / user-overridden values are kept. -
[Reality] Colliding WAFBench case titles no longer emit duplicate k6
constnames (#79). A 32-target--wafbench-verbatimrun with 31 YAML files sharing titles likenormal request allowedfailed every target with k6 exit 107 (ScriptException) and 0 requests. Identifiers and metric names now get a_2,_3, … suffix.
Added
- [Reality] Traffic-file breakdown reprints at end of run with unique vs total (
unique * RPS) and writestraffic-breakdown.jsonunder the output dir for automation (#79).
[0.3.216] - 2026-08-31
Fixed
- [Reality] A missing
--wafbench-dirpath is now a hard error instead of a warning plus an empty payload pool (#79). On 0.3.215,mockforge bench --wafbench-dir missing.yaml --spec ... --targets-file ...still generated a k6 script;getNextSecurityPayload()then returnedundefinedand goja threwCannot convert undefined or null to object. The command now exits withWAFBench path does not exist. Defense in depth: an empty pool returns[]rather thanundefined.
Added
- [Reality] After a traffic file loads, print per-file
sent/attack(expected 403)/normal(expected 200)/omittedcounts so a multi-YAML run has a checklist for proxy-log review (#79).
Security
- [Security] RUSTSEC-2026-0269: bumped
wasmtime36.0.13 to 36.0.14 (filesystem sandbox escape on trailing slashes). Same 36.x line as the rest of the plugin-loader pin.
[0.3.215] - 2026-08-30
Fixed
- [Reality]
--wafbench-verbatimnow works with--targets-file(#79). The single-target path already sent traffic-file requests as written and made--specoptional, but--targets-filedispatched intoParallelExecutorfirst. That path still required a spec and built templates from spec operations, so the same command either died withNo spec files providedor fuzzed spec URLs.ParallelExecutornow loads verbatim templates, keeps--specoptional, and callssecurity_testing_enabled()instead of recomputing the injection flag inline.--base-path /no longer prefixes a second slash onto paths that already start with/.
[0.3.214] - 2026-08-23
Added
-
[Reality] New
mockforge bench --wafbench-verbatimsends each traffic case exactly as written instead of extracting an attack payload from it (#994, #79 round 66 / Srikanth on 0.3.213). His WAF rule chained onARGS:redirect_uriandARGS:response_typenever fired, and the run still went green, which for a security tool is worse than an error. The cause: by default a case’suriis treated as a CRS attack string hidden in a query parameter, soextract_uri_payloadkeeps only the FIRST parameter’s value, discards the path, and re-attaches the survivor to a spec-derived endpoint as?test=<payload>./oauth/authorize?response_type=totally-unsupported&redirect_uri=https%3A%2F%2Fevil...&state=s1therefore went out as roughlyGET <spec-endpoint>?test=totally-unsupported, withredirect_uriabsent entirely. That behaviour is correct for CRS/WAFBench files, where one attack string is the whole test, so this adds a mode rather than changing the default. With the flag, method, full URI, headers and body are sent as given: no extraction, no path substitution, notest=injection, no spec-endpoint cycling, and--specno longer supplies the endpoints. The URI is preserved byte for byte (query params are deliberately not parsed and rejoined, because for charset and traversal cases the encoding IS the payload). Real-binary verified against his own case and his 8-case file. Also documents that--wafbench-cycle-allaffects payload selection only and does not change targeting. -
[Reality]
--wafbench-verbatimnow genuinely sends requests untouched (#997). Found by capturing what the flag actually put on the wire rather than by reading the generated k6 script:--wafbench-dirfeeds two consumers, and while verbatim mode stopped the template from extracting payloads, the security layer still read the same file as a payload POOL and appended&test=<payload>to every request at a later stage. That corrupted the exact cases the user asked to be sent literally, and it attached attack payloads toexpected: 200cases, so a correctly-behaving WAF would block them and the run would report failures the user never wrote. Inventing failures is as damaging as the false passes #994 fixed. The gating flag had been recomputed inline at four render sites, the #79 drift shape, and is now a singlesecurity_testing_enabled()method guarded by a test that fails if inline copies return.--security-testcombined with verbatim now warns and is ignored rather than silently mutating traffic.--specis optional in verbatim mode, since the traffic file supplies every request. -
[Reality] Traffic cases marked
omit_rule: trueare skipped with an explanation instead of being parsed and silently dropped (#997). The field marks a baseline to be sent with the rule under test disabled; mockforge cannot toggle a WAF rule, so the case previously went out byte-identical to its non-omitted twin while asserting the opposite status. Exactly one of the pair always failed regardless of whether the rule worked, which defeats the only purpose of a baseline.
Fixed
- [DevX] Documented the 41
MOCKFORGE_*environment variables that were read in code but absent from user-facing docs, which had left the docs/code drift gate red on every PR regardless of content. Covers the previously undocumented operator surface: plugin host, plugin egress proxy, test runner, platform LLM, capture forwarding, contract-diff and drift body limits, Kafka offset persistence, OTLP port, hosted overage ceiling, PagerDuty endpoint override. Defaults were read from the code rather than guessed, and three easy-to-miss behaviours are called out:MOCKFORGE_PLUGIN_HOST_SIGNATURE_MODEsilently falls back tooptionalon an unrecognised value, andMOCKFORGE_SSRF_ALLOW_LOOPBACK/MOCKFORGE_SSO_ALLOW_INSECURE_ISSUERSrelax SSRF protection and are test-only. - [DevX] Added a WAF Testing page to the book (#79).
--wafbench-dirhad no documentation at all, which is why targeting had to be explained by hand on the issue. Covers the two accepted file shapes, the payload-extraction vs verbatim distinction that decides whether a rule fires, a targeting matrix for the flag combinations, that--wafbench-cycle-allaffects payload selection only, and whyomit_rulecases are skipped.
[0.3.213] - 2026-08-20
Fixed
- [Contracts] OpenAPI specs whose operations omit
responsesnow load instead of aborting the run (#79 round 65 / Srikanth on 0.3.212). His proxy-generated spec failed outright withmissing field \responses`, and the error named no path or method. The spec is genuinely non-conformant (OpenAPI 3.x marksresponsesrequired, and 63 of its 70 operations omit it), but refusing it is the wrong trade: request generation, load testing and conformance probing all derive from paths, parameters andrequestBodyand never readresponses`, so the missing field costs nothing we use while rejecting the document costs the whole run. Proxies and API-discovery tools emit specs like this routinely, because they observe requests and have nothing to say about responses. A narrow repair pass fills in only the mandatory field, never invents response content, and warns with the count of affected operations and what is lost (response-schema validation has nothing to check for them). Real-binary verified against his attachment: the run now loads and finds all 70 operations.
Added
- [Reality]
--wafbench-dirnow accepts a simple{title, request, expected}YAML list in addition to the WAFBench document shape (#987, #79 round 65 / Srikanth). His LLM-generated traffic file was rejected withinvalid type: sequence, expected struct WafBenchFile, which was accurate about what failed but silent about what would have worked. The WAFBench shape exists to carry CRS rule IDs and provenance that a generated file has no reason to contain. The simple form maps 1:1 onto the internal representation (bodybecomesdata, titles are synthesised when absent), so the rest of the security-test pipeline is untouched.expectedaccepts403,[403, 406], or{status: ...}, and an unparsable expectation yields no status rather than dropping the case, since the request is still valid traffic. When both shapes fail, the error now names both with the parse error for each. Real-binary verified against his unmodified file: 8 cases, 24 payloads.
Security
- [Security] RUSTSEC-2026-0258 (
h2unbounded empty DATA frames): bumpedh20.4.15 to 0.4.17, which is the instance that serves HTTP here via hyper 1.x / axum. The remainingh20.3.27 is reached only through the AWS SDK client path and is ignored with reasoning recorded inaudit.toml: 0.3.27 is the latest 0.3.x and upstream patched only the 0.4 line, and the flaw concerns queueing empty frames from a peer, where on that path we are the client talking to AWS KMS/STS over TLS. Advisory is rated low severity.
[0.3.212] - 2026-08-03
Fixed
- [Reality]
--conformance-basic-auth(and--conformance-headers, where--auth-bearerlands) now reach the wire on multi-target runs (#79 round 64 / Srikanth on 0.3.210). Hismockforge bench --use-k6 --targets-file vs_list3.json --conformance-basic-auth user:pass ...sent no credentials at all, absent from both his PCAP and his proxy, while the identical single-target invocation worked. Round 47 taughtparse_headers()to fold those auth shortcuts into the shared header map so the same flags work for a plain load run without--conformance, and both ends of the chain were already correct (ParallelExecutorcallsparse_headers(); the k6 generator mergescustom_headersinto every endpoint). The break was between them:execute_multi_targethand-copiesBenchCommandfield by field and setconformance_basic_auth: None/conformance_headers: vec![], nulling the fields beforeparse_headers()ran, so the fold had nothing to fold. Real-binary verified across two targets:grep -c Authorizationon the generated k6 scripts goes 0,0 to 3,3, carryingBasic dXNlcm5hbWU6cGFzc3dvcmQ=. Adds a drift guard asserting every fieldparse_headers()folds survives that clone, because field-by-field struct literals drop fields silently. Note that--validate-requestsand--export-requestsare conformance-run features (read only insideexecute_conformance_test); they no-op on a plain load run in single- and multi-target alike, and need--conformance. - [Security] RUSTSEC-2026-0222:
wasmtime36.0.12 to 36.0.13, reached viawasmtime-wasi->wiggle->mockforge-plugin-loader. Lockfile only, a patch bump inside the existing 36.x line. - [Contracts]
mockforge-registry-coreno longer fails to compile for default-feature consumers. An associated const added in 0.3.211’s audit fix sat inside a#[cfg(feature = "postgres")]impl while its ungated caller and tests still referenced it, socargo install mockforge-clicould not build. Nothing in CI covered that combination: the workspace test job runs--all-featuresand the registry server always builds withpostgres. Moved to a module-level const. - [Contracts] Repaired the release pipeline itself. A stale doctest in
mockforge-intelligence(broken since May) was failingcargo test --workspace --releasein the Create Release job, which skipped everything after it, including the GitHub Release and the Fly deploy, which hasneeds: release. Production had therefore been serving 0.3.183 since 2026-06-19. Also aligned every Dockerfile builder withrust-toolchain.toml(1.75/1.90/1.91 to 1.96) afteraws-smithy-typesraised its MSRV past the pinned builder and broke container builds outright, and cleared theunused_qualificationserrors that were reddening the Test job on every PR regardless of content.
[0.3.211] - 2026-07-27
Fixed
- [Reality] Target-side HTTP protocol violations now classify as
"kind": "protocol"inconformance-network-events.jsoninstead of the generic"other"(#79 round 63 / Srikanth on 0.3.210). His WAF answered blocked security probes (1234 OR 1=1,waitfor delay, …) with a response body that disagreed with its own declaredContent-Length, which Go’snet/httprejects asserver replied with more than declared Content-Length; truncated. Those 2,824 events were landing in theotherbucket alongside genuinely unclassifiable failures, so a real protocol violation by the target (the class of defect behind request-smuggling and desync bugs) was indistinguishable from noise. The new branch also coversmalformed,protocol error, andinvalid header, and is evaluated last so the existing eof/timeout/tls/connect rules keep their meaning; it only re-labels what would otherwise beother. Applied to all three k6 script-generation paths (generator.rs,spec_driven.rs,k6_script.hbs) with a regression test asserting the branch exists, and stays ordered last, in every path so they cannot drift apart.
[0.3.210] - 2026-07-24
Added
- [Reality] New
mockforge bench --no-abort-on-errorand--abort-on-error-rate <rate>flags to control the round-60 k6 memory safety valve (#79 round 62 / Srikanth on 0.3.209). The round-60 valve aborts a k6 run whosehttp_req_failedrate crosses 0.95 for 60s, which prevented an OOM against dead targets but also stopped legitimate stress tests: Srikanth’s WAF/proxy targets legitimately reject ~95.2% of the generated probe traffic (fast non-2xx responses, zero transport failures), which crossed 0.95 and aborted every run at ~2 minutes.--no-abort-on-errordrops theabortOnFailthreshold entirely so the run executes its full duration regardless of error rate;--abort-on-error-rateretunes the threshold (default 0.95) to, for example, keep the dead-target guard but only fire above 99%. Wired through the single-target and multi-target (--targets-file) k6 paths; default behaviour is unchanged. Real-binary verified: the generated k6 script drops theabortOnFailobject under--no-abort-on-errorand rendersrate<0.99under--abort-on-error-rate 0.99.
[0.3.209] - 2026-07-22
Added
- [Reality] New
mockforge bench --dns-policy <policy>flag that passes k6’s DNS resolution policy through as--dns "policy=<value>"(#79 round 61 / Srikanth on 0.3.208). Values:preferIPv4(k6’s default when unset),preferIPv6,onlyIPv4,onlyIPv6,any. This unblocks GEODB IPv6 load tests where the target must stay a hostname (the proxy routes by Host/SNI, so a bracket IP can’t be used) but the dial has to use the AAAA record: k6/Go default to the A (IPv4) record, which then can’t be dialed from an IPv6--source-ipand fails withno suitable address found.--dns-policy preferIPv6makes k6 pick the AAAA record while keeping the hostname on the wire. Wired through the single-target and multi-target (--targets-file) k6 paths. Real-binary verified:--dns-policy preferIPv6makes the launched k6 child run with--dns policy=preferIPv6in its argv.
[0.3.208] - 2026-07-21
Fixed
- [Reality] k6 no longer gets OOM-killed (
signal: 9 (SIGKILL)) when a load target rejects or times out nearly every request (#79 round 60 / Srikanth on 0.3.207: an IPv6 run where the targets were unreachable ran the full 4200s and k6 climbed to an OOM). The generated k6 load scripts now carry anabortOnFailsafety valve onhttp_req_failed: if a target fails>=95%of requests after a 60s grace period, that target’s k6 run aborts (exit 99) instead of hammering a dead endpoint for the whole duration and accumulating per-request metric/error memory. Legitimate runs, even heavily degraded ones under 95% failure, are unaffected (the existing pass/fail error threshold is unchanged). Real-binary verified against a dead target (connection refused, 100% failure): k6 now stops at ~60s with “thresholds … abortOnFail enabled, stopping test prematurely” instead of running the full 180s scenario. Applies to both the standard load and CRUD-flow k6 templates. Separately, the accompanying diagnosis on #79 covers the IPv6 dial failures behind the OOM (source-IP address-family mismatch on hostname targets, TLS/connection refusals on the:443targets), which are network/target-side, not memory.
[0.3.207] - 2026-07-19
Added
- [Contracts] New “Security probes (owasp injection)” view in the conformance self-test (#79 round 59 / Srikanth on 0.3.206 was testing a WAF and saw
owasp: 0 caught / 8127 missedbutDefinite issues: none, and asked whether real issues were being hidden). The self-test now prints a per-injection-family breakdown of how many OWASP payloads the target accepted (status < 400) vs blocked (4xx), and writes aconformance-owasp-accepted.jsonsidecar listing every accepted payload with its method + URL. For a WAF / security proxy, each accepted payload is one it did NOT block; for a plain API this is expected (astringfield accepts any string, so an SQLi string is spec-valid). It is deliberately kept out of “Definite issues” (which is for spec violations), but surfaced on its own line so a security tester sees it without eyeballing the caught/missed rollup. Verified against Srikanth’s own capture: the summary reports the target accepted all 8127 injection payloads across all 7 families (sqli / xss / command-injection / path-traversal / ssti / ldap-injection / xxe), and real-binary verified againstmockforge servethat the section and sidecar render.
[0.3.206] - 2026-07-16
Fixed
- [Reality] Multi-target load no longer fails to start with
k6 exited with status: exit status: 106and zero requests (#79 round 58 / Srikanth on 0.3.205: all 10 targets failed this way, so no traffic ran and--discard-response-bodiesappeared to “not help”). Exit 106 is k6’sCannotStartRESTAPI: each parallel k6 was pinned to a FIXED REST API port (6565 + index→ 6566, 6567, …), which collides whenever those ports are already taken (orphaned k6 from a previous OOM-killed run, a second concurrent bench, or any local service). Each k6 instance now binds an OS-assigned ephemeral port (--address localhost:0) for its REST API, so it can never collide; MockForge never queries that API (results come fromsummary.jsonon disk). Real-binary verified: with ports 6566 and 6567 deliberately occupied, a multi-target load that previously exited 106 on every target now starts cleanly and generates traffic on all targets.
Added
- [Contracts] New “Definite issues” view in the conformance self-test (#79 round 58 / Srikanth on 0.3.205 asked for a way to say “for sure this is an issue” without eyeballing every caught/missed row across all APIs). A
Definite issues (N)section is printed under theNegatives [...]summary and aconformance-definite-issues.jsonsidecar is written next to the report. It surfaces only the unambiguous problems that need no per-probe analysis: (1) a spec-valid request the target rejected, and (2) any probe that drew a 5xx (the target crashed instead of replying cleanly). Spec-valid “missed” negatives (which genuinely need judgement) are deliberately excluded. Real-binary verified: a clean target prints “Definite issues: none”; a target that 404s a valid request surfaces it as avalid_request_rejectedrow in both the console and the JSON. Also clarified in the docs that a “caught” 4xx is the target doing its job (rejecting a bad request), not a violation.
[0.3.205] - 2026-07-14
Added
- [DevX] New
mockforge bench --discard-response-bodiesflag exposesK6_DISCARD_RESPONSE_BODIES=trueas a first-class option (#79 round 57 / Srikanth on 0.3.204 asked for a flag instead of setting the env var for scale/stress runs). Plain load runs only check status codes but k6 buffers every response body by default, which OOM-kills k6 on long, high-VU runs. The flag opts a single-target load run into discarding bodies (multi-target load already discards by default since 0.3.204); it is a no-op on conformance / self-test / CRUD-extraction runs, which need the body. Verified with the real binary: passing the flag makes the launched k6 child carryK6_DISCARD_RESPONSE_BODIES=trueand keepsdata_receivedat header-only levels, while omitting it leaves the env unset.
Changed
- [DevX] The self-test console summary now prints a one-line legend under the
Negatives [...]lines explaining that “caught” means the target rejected a deliberately-bad request with a 4xx and “missed” means the target accepted it, and that a high “missed” is not automatically a bug because many probes are spec-valid by construction (#79 round 57 / Srikanth on 0.3.204 asked how to read caught vs missed). The self-test-probes reference page gains the same spec-valid nuance (astringfield accepts any string so anowasp:sqlipayload is a valid string; an optional query param can be dropped; an unconstrained{id}accepts any value), and the capacity-sizing page documents the new flag.
[0.3.204] - 2026-07-13
Fixed
- [Contracts]
parameters:*negative probes are now recorded inconformance-request-violations.json, and the single-target self-test writes that file at all (#79 round 56 / Srikanth on 0.3.203: parameter violations were still absent from the logs). Three compounding gaps: (1) for his Apigee spec the parameter probes are all spec-VALID by construction (dropping the OPTIONAL$.xgafvquery, adding an oversized/undeclared query param, or an unconstrained{instance}path value violate no schema), so the emitted-request validator found nothing to flag and stayed silent; (2) the SINGLE-target self-test returned before ever calling the validator (only the r49 multi-target--targets-filepath invoked it); (3)--validate-requestsdid not enable request capture, so even when the validator ran it had no emitted requests to walk. Now eachparameters:*negative that produces no hard breach is recorded as a typedparameter_negative_proberow with a message explaining why it is not a schema violation (and that rejection is the target’s responsibility); the single-target self-test calls the validator too; and--validate-requestsimplies capture. Real-binary verified againstmockforge servewith a custom-verb Apigee spec: the violations file now contains all three parameter probes (uri-too-long,bad-path-param,missing-query) alongside the genuinebody_schema_violation/query_value_mismatchrows, in both single-target and--targets-filemodes. - [Reality] k6 no longer gets OOM-killed (
signal: 9 (SIGKILL)) during long, high-concurrency multi-target load runs (#79 round 56 / Srikanth on 0.3.203: a ~70-minute run across 10+ targets at 10 VUs each was killed mid-run). The generated plain-load k6 only checks status codes but was buffering every response body in memory; at that VU count over that duration the buffered bodies plus k6’s own metric accumulation exhausted RAM and the kernel OOM-killer sent SIGKILL. The multi-target plain-load path now runs k6 withK6_DISCARD_RESPONSE_BODIES=true(safe there because bodies are never inspected). Real-binary verified: the launched k6 child process carriesK6_DISCARD_RESPONSE_BODIES=truein its environment anddata_receivedstays at header-only levels. Body-inspecting paths (conformance, extraction) are untouched; the flag defaults off and is opt-in per executor.
[0.3.203] - 2026-07-12
Fixed
- [Contracts]
--base-path /no longer voids all conformance validation (#79 round 55 / Srikanth on 0.3.202: a self-test with--base-path /wrote an empty violation file even though the console reported thousands of owasp/parameter/request-body misses). Root cause: the self-test URL builder computed the prefix as"/".trim_end_matches('/')="", decided"".starts_with('/')was false, and produced the prefix"/"; prepended to the already-rooted spec path that yielded<target>//v1/organizations(double slash), whose template//v1/organizationsnever matched the spec’s/v1/organizations, so the validator skipped every emitted request and logged nothing (even the OWASP/request-body probes that worked in prior rounds). A sharedbase_path_prefix()now trims slashes from BOTH ends and treats the empty result (/,//, ``) as no prefix (root), used by both prefix sites; the validator also collapses stray double slashes before matching. Real-binary verified with Srikanth’s exact command: the emitted URL is now.../v1/organizations(single slash) and the logs populate again (7 owasp + 6 request-body), where 0.3.202 wrote[].
[0.3.202] - 2026-07-11
Fixed
- [Contracts] Parameter violations now surface in the by-request / by-probe logs, not just OWASP and request-body ones (#79 round 54 / Srikanth on 0.3.200:
parametersnegatives were missed in the console but absent from the violation files). Two gaps: (1) aparameters:bad-path-paramprobe targets a custom-verb path like/v1/{instance}:reportStatus, whose templated segment{instance}:reportStatusnever matched the path template, so the whole request was skipped before any parameter check ran; (2) parameter value validation only covered enum/integer/number/boolean, not stringpattern/minLength/maxLength. A new segment matcher binds{name},{name}:verb, andprefix{name}forms (used by both the path matcher and the path-param extractor), and string length/pattern constraints are now checked. Real-binary verified: a self-test against a spec with a custom-verb path and amaxLength-constrained param now writesparameters:bad-path-param => path.instance: value "self-test-invalid-id" is longer than maxLength 8. (Probes that drop an OPTIONAL query param or add an extra param stay spec-valid by construction and correctly produce no client-side violation; the full negative-probe record is inconformance-self-test.json.) - [Reality]
--source-ipnow takes effect with--targets-file(#79 round 54 / Srikanth on 0.3.200: source-ip was a silent no-op in multi-target mode). The per-targetBenchCommandclone inexecute_multi_targethardcodedsource_ips: Vec::new(), and theParallelExecutorbuiltK6Executor::new()without.with_local_ips. The source IPs (and geo variants) now flow through the clone and--local-ipsis passed to each target’s k6. Real-binary verified:bench --targets-file --source-ip 10.99.99.99now makes k6 fail withbind: cannot assign requested address(proving the source IP is applied), where before it connected from the default address.
[0.3.201] - 2026-07-10
Added
- [Reality] New
mockforge bench-qoscommand: a native QoS / DSCP traffic-class load generator (#933 / Srikanth on #79 asked whether bench can emit Voice / Video / Background / Best-Effort traffic in one run to exercise a path’s QoS handling). k6 is an application-layer client with no L3/L4 knobs, so this uses rawsocket2sockets and sets the IPv4IP_TOS(DSCP) byte on each TCP connection BEFORE connect, so the SYN and every segment carry the marking. Presetsvoice(EF/46 → 0xB8),video(AF41/34 → 0x88),best-effort(0x00),background(CS1/8 → 0x20), ordscpNNfor a raw code point;--class NAME[:WEIGHT]is repeatable and mixes classes by relative weight in a single run.--mssclampsTCP_MAXSEGto approximate small-MTU behavior. Prints a per-class table (dscp, tos byte, requests, ok, p50/p95) and warns if the kernel rejected the marking. Jumbo frames and IP fragmentation are NIC/kernel-level (not socket options), so the command prints theip link/tc netemrecipes for those; IPv6IPV6_TCLASSis a follow-up. Verified: a unit test readsIP_TOSback with getsockopt off the exact generator socket for every preset (the kernel stamps that into outbound IP headers, so it’s the wire guarantee), and a real run sent 62,115 successful DSCP-marked requests split across classes, with weighted mixing and the custom-DSCP / MSS options confirmed.
[0.3.200] - 2026-07-10
Fixed
- [Contracts] Emitted-request validation now reports
owasp:*and parameter violations, not just request-body ones (#79 round 53 / Srikanth on 0.3.199: the by-request / by-probe logs showed onlyrequest-body:*rows even though the console reported 7602 missed owasp and 3248 missed parameter negatives). Two causes: (1) every OWASP probe injects into$.xgafv, which reaches the wire percent-encoded as%24.xgafv; the validator built its query map from the raw query string then looked each parameter up by the spec’s DECODED name, so the lookup missed and all 7602 probes were skipped silently. Query keys/values and path segments are now percent-decoded (and$refparameter schemas resolved). (2) Aparameters:missing-queryprobe that drops a REQUIRED parameter now surfaces asquery.<name>: required parameter missing; before, the loop only inspected parameters that were actually sent. Real-binary verified againstmockforge serve: all seven OWASP checks now appear as e.g.owasp:sqli => query.$.xgafv: value "' OR '1'='1" is not one of "1" or "2". - [Reality] PUT (and any) request bodies that are not JSON are no longer rejected as missing (#925: every
PUTwith anapplication/octet-stream/ XML / urlencoded body returned400 body: Request body is required but not provided, while JSON POSTs passed). The handlers derived body presence fromserde_json::from_slice(&bytes).ok(), so a non-JSON body collapsed toNoneand looked absent. Body presence is now tracked from the raw bytes,requestBody.requiredis honored, and the AI handler no longer usesOption<Json<Value>>(which axum yields asNonefor non-JSON). Real-binary verified: PUT octet-stream -> 200, empty body againstrequired: true-> 400, JSON schema validation unchanged. - [DevX] A relative
openapi_specconfig value now resolves from the config file’s directory when it does not resolve from the working directory (#928: in Docker, a relative path resolved from/app(CWD) instead of/configand silently failed to load). Absolute and CWD-resolvable paths are untouched. Verified from a foreign CWD. - [DevX]
servefails fast when a configured OpenAPI spec cannot be loaded, instead of starting “healthy” with zero routes (#929). Clear error names the path; setMOCKFORGE_ALLOW_MISSING_SPEC=1to start anyway. Verified against a config-driven bad path. - [DevX]
http.request_validationis honored as a deprecated alias forhttp.validation.mode, and unknownhttp:config keys now log a warning instead of being dropped silently (#927). - [DevX] The Docker image now uses an entrypoint shim so the documented
command: serve --config ...works (#930: the image only set CMD, so a usercommand:execedserveas a binary). The oldcommand: mockforge serve ...workaround still works. - [DevX] The container user now has a real, owned home directory, so
~/.mockforge/scenariosis writable and the startup “Failed to create scenarios directory: Permission denied” warning is gone (#931). - [Cloud] Bumped
crossbeam-epochto 0.9.20 (RUSTSEC-2026-0204, invalid pointer dereference) andquinn-prototo 0.11.15 (RUSTSEC-2026-0185/-0037); both were transitive. Merged the dependabot cargo patch/minor group (14 updates incl.rustls0.23.41,anyhow1.0.103,tauri2.11.5) and the GitHub Actions bump group.
Note
- [Contracts] #926 (stale Content-Type allowlist after a spec change) is addressed by the #928 / #929 fixes: MockForge reads the allowlist fresh from the loaded spec per request with no static cache, so the reported persistence was the relative-path failure silently loading a stale spec. Srikanth’s QoS / L3-L4 traffic-generation request is tracked in #933.
[0.3.199] - 2026-07-06
Fixed
- [Contracts] Emitted-request validation now populates the by-request / by-probe violation logs for specs whose request bodies are
$refs (#79 round 52 / Srikanth on 0.3.198:--conformance-self-test --targets-filereported thousands of caught request-body negatives but wrote emptyconformance-request-violations-by-request.json/-by-probe.json). Root cause: the client-side body check only fired when the operation’srequestBodymedia schema was an inlineReferenceOr::Item; the Apigee spec (like most real specs) declaresschema.$ref = #/components/schemas/..., soschema_ref.as_item()returnedNoneand every body probe was silently skipped. It also only type-checked STRING values, so a{"analyticsRegion":12345}(number-where-string) probe surfaced nothing even when the schema resolved. The validator now resolves therequestBodyand schema$refs and reuses the same full JSON-Schema validator the custom-checks path uses (build_validator), so nested$refs, root-type mismatches, enum violations, and non-string type mismatches all populate the logs. Real-binary verified: a self-test against a$ref-bodied spec now writes 16 flat / 8 by-request / 15 by-probe violations (previously empty), e.g.body/analyticsRegion: 12345 is not of type "string",body$: [] is not of type "object". This also makes the OWASP probes’ query-vs-body attribution visible (e.g.owasp:sqli => query.alt: value "%27%20OR..." is not one of "json" or "media").
Added
- [Contracts] Multipart upload summary now reports the full HTTP request size alongside the entity body, so a proxy’s byte count reconciles (#79 round 52 / Srikanth on 0.3.198: a persistent 264-byte gap between his proxy’s
57998271and MockForge’s57998007). MockForge’swirefigure is the exact RFC 7578Content-Length; the 264-byte delta is the top-level HTTP request preface (request-line +Host+ the script-visible headers + the transport-managedContent-Length+ the blank line) that a proxy meters but MockForge previously did not surface. The k6 summary now also emitsrequest N bytes incl H-byte header block. Verified against a live raw-socket capture: in one k6 run the reportedwire/request/header-block (4838 / 5053 / 215) equalled the bytes k6 actually put on the socket, to the byte. Fixed in both render paths (generator.rs+spec_driven.rs). - [AI] The mock MCP server ships a richer default catalog so clients generate more capturable traffic (#79 round 52 / Srikanth on 0.3.198: “Is it possible from mockforge mcp mock server to respond with more tools, resources, prompts so that applications are chatty”). The default
--mcp-mockcatalog now advertises 8 tools (echo,get_status,get_time,add,search_documents,get_weather,create_ticket,fail_tool), 4 resources, and 3 prompts, and implementsresources/read,prompts/get, andresources/templates/list(previouslyresources/listandprompts/listwere empty and onlytools/listreturned content). A client that lists then reads/gets/calls now exchanges many more request/response pairs. Unknown resource uri / prompt name return a JSON-RPCINVALID_PARAMSerror. Real-binary verified againstmockforge serve --mcp-mock.
[0.3.198] - 2026-07-02
Fixed
- [Contracts] Self-test probes no longer trip spurious query/body violations that get attributed to the wrong check (#79 round 51 / Srikanth on 0.3.196: a
request-body:content-type-mismatch:multipartprobe reportedquery_value_mismatchonalt/prettyPrint, “incorrect and confusing”). Root cause: the baseline probe filled every query/path param with the literaltest-valueand every string body property withtest-string, both invalid against enum/boolean/type constraints, so every probe tripped the same incidental violations regardless of what it was testing. The baseline filler is now spec-VALID: enum -> first member, boolean ->true, integer/number -> a valid number, string formats (email/uuid/date/…) -> a plausible value, and enum body properties -> a valid member. Negative probes still override the specific param/body they attack. Real-binary verified againstmockforge serve: on the Apigee-style spec the by-probe violation count dropped from 32 to 12, everyrequest-body:*probe now showsquery_viols=0, and thecontent-type-mismatchprobes are clean. (This also makes the SQLi/OWASP probes truthful: they now show the violation for the param they actually injected, not unrelated filler noise.) - [Contracts] Multipart
wirebyte count now matches a proxy’s observed body length exactly (#79 round 51 / Srikanth on 0.3.196: the r50 reconstructed envelope was 264 bytes short of his proxy’s 1955). The k6-side reconstruction fromres.request.bodycan drift from the real socket bytes, sowirenow prefers the request’sContent-Lengthheader (the exact wire body size k6 puts on the socket, which is what a proxy counts), falling back to the reconstructed envelope only when the header is absent. Fixed in both render paths (generator.rs+spec_driven.rs).
Added
- [AI][DevX] New reference page “Agent / LLM / MCP Traffic (Packet-Level)” (#79 / Srikanth on 0.3.196 asked for a document explaining Agent<->LLM, Agent<->Agent, and MCP-Agent<->MCP-Server interactions at the HTTP-packet level with PCAP guidance). Documents the wire structure (methods, headers, request/response bodies, SSE streaming, JSON-RPC envelopes) for OpenAI, Anthropic, and MCP using real captures taken against
mockforge serve --llm-mock --mcp-mock, plus copy-pastetcpdump/tshark/mitmproxy/SSLKEYLOGFILErecipes for recording true.pcapfiles.
[0.3.197] - 2026-07-01
Added
- [AI] Mock LLM endpoint gains real-model modes via
--llm-mock-mode(#915, follow-up to #912). The default stays a pure offlinemock, and three new opt-in modes let it work with a real OpenAI/Anthropic-compatible upstream:proxyforwards every request to--llm-mock-upstreamand returns the real response (combine with--latency/--failuresfor chaos on live traffic);recordforwards on a cassette miss and saves the response to--llm-mock-cassette, replaying from the cassette on a hit (real content, deterministic after warm-up);replayserves only from the cassette (fully offline), falling back to the canned reply on a miss. New flags:--llm-mock-mode,--llm-mock-upstream,--llm-mock-api-key(or$OPENAI_API_KEY/$ANTHROPIC_API_KEY),--llm-mock-cassette. Upstream/cassette failures degrade to the canned reply so the endpoint never hard-fails a caller. Reuses the existing reqwest client; upstream calls forward the model + messages verbatim (always non-streaming) and streaming clients get the resolved text re-chunked locally. Real-binary verified againstmockforge serveby pointing one--llm-mockinstance at another as the upstream (no API key needed): record captured to the cassette then replayed with the upstream stopped; purereplayserved hits offline and fell back to canned on a miss;proxylive-forwarded both OpenAI and Anthropic dialects.
[0.3.196] - 2026-06-30
Added
- [AI] Mock LLM endpoint for agent testing (#912 / #79 Srikanth: “Agent Traffic to LLM … LLM Simulation using MockForge Server”).
mockforge serve --llm-mockmounts OpenAI-compatiblePOST /v1/chat/completions+GET /v1/modelsand Anthropic-compatiblePOST /v1/messages, so an agent (Cursor, Claude Code, ChatGPT clients, custom agents) can point its base URL at MockForge and receive correctly-shaped, deterministic completions. Responses carry realistic envelopes (ids,usage/token counts,finish_reason/stop_reason) and SSE streaming when the caller setsstream: true(OpenAIchat.completion.chunkdeltas +[DONE]; Anthropicmessage_start/content_block_delta/message_stopevents). Pure mock (no external model call); echoes the prompt by default so wiring is easy to confirm. Real-binary verified againstmockforge serve: non-streaming + streaming OpenAI, Anthropic messages, and model list all return spec-shaped payloads. - [AI] Mock MCP (Model Context Protocol) server for agent-as-MCP-client testing (#913 / #79 Srikanth: “Agent (Acting as MCP Client) Traffic to MCP Servers”).
mockforge serve --mcp-mockmounts a JSON-RPC 2.0 endpoint atPOST /mcpansweringinitialize,tools/list,tools/call,resources/list,prompts/list, and thenotifications/initializednotification, with a configurable tool catalog returning canned (or, via an error tool, failing) results. Real-binary verified againstmockforge serve:initializereturnsserverInfo+ capabilities,tools/listreturns the catalog, andtools/call echoreflects its argument.
[0.3.195] - 2026-06-30
Fixed
- [Contracts] Multipart
wirebyte count no longer comes out SMALLER thantotal(#79 round 50 / Srikanth on 0.3.194: “wire 56344432 bytes … is less than file size bytes in disk 57996316”). r49 setwire = raw.length, but k6’sres.request.bodyis a UTF-8-decoded JS string, so binary part bodies collapse andraw.lengthUNDERcounts (his 58 MB upload read ~1.65 MB low). The multipart envelope is only ~2 KB for 9 parts, so wire can never be less than total. Now wire is computed as the disk-accurate payload sum plus the ASCII envelope (each part:--boundary+ part headers + the header/body CRLFs + trailing CRLF, then the closing--boundary--CRLF) — all ASCII, so byte-accurate even when binary bodies mangleraw.length. Fixed identically in both render paths (generator.rs+spec_driven.rs). Verified by running the emitted k6 JS in node against a body whoseraw.lengthundercounts:wire = total + envelopeandwire > total. - [Contracts]
conformance-request-violations-by-request.jsonnow emits exactly ONE row per URL (#79 round 50 / Srikanth on 0.3.194: a check seen in by-probe for a URL — e.g.owasp:ldap-injection— was missing from the by-request row for that same URL). The r47 grouper keyed on(method, path, violation-set), so a URL with both query/owasp violations and request-body violations split into separate rows and his “find the URL with the most violations, then drill into by-probe” flow only saw one half. Now each(method, path)collapses into a single row carrying the deduped UNION of every violation and every contributingcheck_name; per-probe attribution stays in the by-probe sidecar. Real-binary verified againstmockforge serve: 0 by-request rows miss any check the by-probe file lists for the same URL, and every by-probe URL has a by-request row. - [Contracts] Repeated self-test iterations no longer duplicate violations in the logs (#79 round 50 / Srikanth on 0.3.194: “I see same violation is getting printed in logs for 22 times. Can this be optimized to print 1 time”). A multi-iteration run captures one probe per iteration, so the flat violation list held N copies of every violation and the grouped rows repeated each
violation_N. The flat list is now deduped by(check_name, method, path, violation_type, message), and both grouped views dedup identical violations within a row. Real-binary verified againstmockforge servewith--conformance-self-test-iterations 3: the flat file collapsed 246 → 82 entries, 0 grouped rows carry a duplicate violation.
Added
- [Contracts] New security-payload categories
llm-prompt-injectionanddlpfor the--security-categoriesflag (#79 round 50 / Srikanth on 0.3.194: “Does mockforge support … Agent simulating DLP (credit card, SSN etc), OWASP Top 10 AI Attacks … via Agent”).llm-prompt-injection(OWASP LLM01; aliasesprompt-injection,llm01,jailbreak) injects inert prompt-override / jailbreak / chat-template-delimiter / context-exfiltration strings into a target field so an agent (k6 VU) can exercise an LLM-or-LLM-backed-API guardrail.dlp(aliasespii,sensitive-data,data-leakage) injects synthetic, documented non-real PII canaries (the 4111… Visa test PAN, reserved000-00-0000SSN, AWSAKIA…EXAMPLEkey shape, PEM private-key shape, reserved-555 phone) for data-loss-prevention egress testing. Both compose with the existing URI/header/body injection and high-risk filtering. Real-binary verified:--security-test --security-categories "llm-prompt-injection,dlp"generates a k6 script carrying the new payloads.
[0.3.194] - 2026-06-29
Fixed
- [Contracts] Violation logs (flat + by-request + by-probe) now write per target in multi-target self-test mode (#79 round 49 / Srikanth on 0.3.193: “I am not seeing any violation requests logs when running [self-test + –targets-file]”). Root cause:
validate_emitted_requestsonly readconformance-requests.json(the bench export shape), but self-test writes theCaseCaptureshape toconformance-self-test-requests.jsonlinstead. Validator now reads whichever file exists, converting JSONL entries into the{check, method, url, request.body}shape the rest of the pipeline expects. Plusexecute_multi_target_self_testcallsvalidate_emitted_requests_with_base_pathper target after the JSONL is written. Real-binary verified againstmockforge servewith self-test +--targets-file: each target now writesconformance-request-violations{,-by-request,-by-probe}.jsonnext to the JSONL. - [Contracts]
--conformance-self-test-durationno longer overshoots by a full iteration (#79 round 49 / Srikanth on 0.3.193: “the test ran more than 5 min (or 46 seconds more)”). The round-47 outer loop only checked the deadline AFTER each matrix pass completed, so a 5m budget on a 6m matrix ran 6m total. Newrun_self_test_with_deadlineaccepts an absoluteInstant; the per-operation loop checks it before each probe and breaks out mid-iteration when the budget elapses. Real-binary verified: 5s budget with--conformance-delay 200ran exactly 5.1s (was minutes of overshoot before).
Added
- [Contracts] Every captured self-test probe now carries an
iterationcounter (#79 round 49 / Srikanth on 0.3.193: “Is it possible to differentiate in the logs what is the iteration count that way I will know how many requests are sent with that violation”).SelfTestConfig.current_iterationis stamped onCaseCaptureso the JSONL line carries the loop counter (1-indexed). command.rs bumps it before eachrun_self_test_with_deadlinecall. Defaults to 1 for older clients / single-iteration runs so existing JSONL deserialises unchanged. - [Contracts] Multipart summary now surfaces both
total(disk-sum payload) andwire(total + multipart envelope) byte counts (#79 round 49 / Srikanth on 0.3.193: “Our Proxy shows Total Byte Length as 57998271 Bytes, Where as mockforge shows … 57996316 Bytes which is correct. I see a diff of 1955 bytes”). The diff is the multipart envelope (boundary lines, per-partContent-Disposition/Content-Typeheaders, CRLFs, final closing boundary), not a bug. New format:<multipart/form-data; boundary=...; N part(s); total <disk-sum> bytes (wire <disk-sum + envelope> bytes w/ envelope): ...>. Real-binary verified with a 256000 + 18 byte upload pair:total 256018 bytes (wire 256434 bytes w/ envelope). The 416-byte delta matches the actual k6-emitted envelope size.
[0.3.193] - 2026-06-28
Fixed
- [Contracts] Multipart upload total bytes now match the sum of per-file disk sizes, not the JS string length (#79 round 48 / Srikanth on 0.3.192: “Only the Total sum of bytes is little incorrect, It should be 57996316 but I see 56344432”). Per-file counts were already exact from r47’s
__mfUploadSizesmap; thetotalBytesline still usedraw.length(UTF-16 code units). NowtotalBytesis recomputed asparts.reduce((acc, p) => acc + p.bytes, 0)after the per-file overlay, only when every part’s size came from disk; falls back to raw.length when any size is unknown. Real-binary verified againstmockforge servewith a 256 KB binary + 18 byte JSON pair: total reports256018 bytes, exactly matchingdisk_size(binary) + disk_size(json). - [Contracts]
--conformance-self-test --conformance-self-test-iterations Nand--conformance-self-test-duration <DUR>now actually loop when paired with--targets-file(#79 round 48 / Srikanth on 0.3.192: “I ran following commands to test conformance-sef-test duration, but the test ended immediately” — both iterations=10 and duration=5m did exactly one matrix pass). Root cause: the multi-target dispatch returned intoexecute_multi_target_conformance(regular k6 flow), never reaching the r47 self-test iteration loop. Newexecute_multi_target_self_testruns the self-test driver against every target with the same iteration / duration logic and writes per-target JSONL, JSON report, andconformance-network-events.json. Real-binary verified againstmockforge servewith 2 targets and--iterations 3: each target ran exactly 3 iterations (positives 0/3, negatives 0/21 + 0/6 + 0/48 per target), per-target reports written undertarget_0/andtarget_1/. - [Contracts] Plain
mockforge bench --use-k6(no--conformance) now writesconformance-network-events.json(#79 round 48 / Srikanth on 0.3.192: “I now see the file … which has no failure info even though I see 18 request failed due to multiple network operations”). The r47 marker only existed in the conformance generator; the plain-bench Handlebars template skipped it. Now emitsMOCKFORGE_NETWORK_EVENT:wheneverres.status === 0directly inside thek6_script.hbsrequest loop, with the same classification logic the conformance + self-test paths use (connect / timeout / tls / otherplus an EOF-detection branch that classifies EOF asconnectsince k6’sRequest Failed error=EOFis a mid-stream connection drop). The CLI executor already harvests the marker. Real-binary verified: plain k6 bench against a closed port writesconformance-network-events.jsonwith one entry,kind: "connect",error_code: 1212,message: "dial: connection refused". - [Contracts] Grouped violation rows now pick a representative
check_namethat matches the dominant violation type (#79 round 48 / Srikanth on 0.3.192: “In the check_name I see body:json but for the violation it is showing some issue with query parameters. Little misleading”). The round-46 grouper returned the alphabetically first contributing check. New logic inspects the row’s first violation_type and picks the contributing check whose name prefix matches (e.g.query_value_mismatch→ prefer aparam:query:*check). Falls back to first when nothing matches. Real-binary verified: a query-only spec now reportscheck_name: "param:query:string:/orgs"for query violations instead of the alphabetically firstconstraint:enum:/orgs.
Added
- [Contracts] New
conformance-request-violations-by-probe.jsonsidecar emits one row per(check_name, method, path)with its full violation list (#79 round 48 / Srikanth on 0.3.192: “Can I assume all this checks has some violation either in the incoming request or outgoing response if yes then how can I see all this violation individually? Do we have any other Logs pointing each of those so that I can fix in one go?”). Unlike the by-request file (which collapses identical violation sets across check names), the by-probe file preserves the per-probe view so the user sees WHICH spec-probing pattern (body:json,schema:string,constraint:enum,method:POST, etc.) surfaced WHICH violation. Sorted by(method, path, check_name)so probes against the same endpoint group visually. Real-binary verified: 30 flat violations expand to 10 by-probe rows (one per contributing check), versus 1 collapsed row in the by-request file.
[0.3.192] - 2026-06-27
Fixed
- [Contracts] Multipart upload byte counts now come from the on-disk file size, not the k6 JS string length (#79 round 47 / Srikanth on 0.3.191: “file bytes are incorrect” with 8/9 file sizes mismatching by ~3-4%). Root cause: k6’s
res.request.bodylossy-decodes binary bytes into a UTF-16 string where multi-byte UTF-8 sequences collapse to single code units, sopartBody.lengthundercounted. Fix: at script-gen timestd::fs::metadata()reads each upload’s actual size into a per-check JS map (globalThis.__mfUploadSizes[checkName][fieldName]) at init scope; the captureExchange summariser overlays those exact counts onto the boundary-derived parts list when it builds the body summary. Real-binary verified againstmockforge servewith a 256 KB binary file + a 18 byte JSON file: export reads'binary':'binary-test.bin' (application/octet-stream, 256000 bytes), 'text':'json.json' (application/json, 18 bytes), matching disk exactly. - [Contracts]
--auth-bearer/--conformance-basic-authnow fire on plainmockforge benchruns too, not just under--conformance(#79 round 47 / Srikanth on 0.3.191: “I tried following command it did not work for bench” — same--conformance-basic-auth user:passhe used on 0.3.190 with conformance was being silently dropped on a non-conformance bench). Extendedparse_headers()to fold the conformance auth shortcuts (--conformance-basic-auth, the round-46--auth-bearercollected viaconformance_headers) into the shared header map after parsing--headerentries; explicit--header "Authorization: ..."still wins.--conformance-api-keystays conformance-only (its semantics are probe-style, not header-style), with a one-line warning when set without--conformance. Real-binary verified:mockforge bench --conformance-basic-auth alice:s3cr3t --use-k6 ...producesAuthorization: Basic YWxpY2U6czNjcjN0in every emitted request; same for--auth-bearer eyJhbGc...→Authorization: Bearer eyJhbGc.... - [Contracts]
conformance-network-events.jsonnow populates under both--conformance-self-testand--use-k6paths, not just the round-46 native executor (#79 round 47 / Srikanth on 0.3.191: “I did not see network logs in the mockforge bench and conformance traffic”). For self-test, added anetwork_events: Option<Arc<Mutex<Vec<NetworkEvent>>>>sink onSelfTestConfig; everyreqwest::Errorfromsend().awaitis classified intoconnect / timeout / tls / body / decode / request / otherand pushed, then drained to disk at run end (writes an empty array on a clean run, the cleanest signal that connectivity stayed up). For k6, the captureExchange JS now emits a separateMOCKFORGE_NETWORK_EVENT:log line wheneverres.status === 0, decoding k6’serror_coderanges (1200s=connect, 1300s=tls, 1400s=timeout); the CLI executor harvests that marker into the same file. Real-binary verified pointing self-test athttp://127.0.0.1:1: 26 wire-levelconnectevents captured with RFC-3339 timestamps. Same target via--use-k6: 10 events, every one carryingerror_code: 1212, kindconnect. - [Contracts] Round-46 grouped violations file no longer shows the same violation set under three different check names (#79 round 47 / Srikanth on 0.3.191: “Only thing is we have same violations on 3 different check name, not clear as to what is the difference”). The grouper now keys on
(method, url, sorted-violation-set)and lists every contributing probe in achecks: [...]array. Real-binary verified againstmockforge servewith a 3-violation POST: 30 flat per-violation entries collapse from 10 grouped rows down to 1 row carrying all 10 contributing check names inchecks: [...].
Added
- [Contracts] New
categoriesarray onServerConformanceViolationlists EVERY distinct location the validator’sdetails[]payload surfaced, not just the priority winner (#79 round 47 / Srikanth on 0.3.191: “I will not know other violations until previous ones are fixed. Is it possible to give a option to show all the violation irrespective of the order”).categorykeeps its single-location shape for dashboards that filter by one category;categories: ["query","request-body"]shows everything that’s broken in one place so the user can address all violations in parallel. Older clients without the field deserialise as empty and fall back tocategory. Real-binary verified: a POST with both query enum and body enum violations returnscategory: "query"ANDcategories: ["query", "request-body"]. - [Contracts]
--conformance-self-test-iterations Nand--conformance-self-test-duration <DUR>knobs let the self-test probe matrix loop (#79 round 47 / Srikanth on 0.3.191: “Is there a way I can run conformance-self-test for longer duration or the duration I specify to simulate and test all the network related problems. Currently for some of the specs it runs and finishes quickly cant even try to do any network triggers”). Iterations defaults to 1; duration accepts the same shorthand as--duration(120s,5m,1h). When both are set, duration wins (iterations becomes the floor). Per-category caught / missed counters sum across iterations via a newSelfTestReport::merge_iterationhelper. Real-binary verified:--conformance-self-test-iterations 3runs every probe three times (positives 0/3, negatives 21/0/6/0/48/0 — exactly 3× the single-iter counts);--conformance-self-test-duration 6sruns for 6 seconds wall-clock then exits.
[0.3.191] - 2026-06-26
Fixed
- [Contracts] Multipart upload no longer lands
[]inconformance-requests.json(#79 round 46 / Srikanth on 0.3.190: “I now see blank entry in conformance-requests.json”). Two stacked fixes. (1) Theextract_exchange_jsonparser used a naivereplace("\\\\","\\").replace("\\\"","\"")chain that mishandled the\\\\\"sequences logfmt emits when a multipart body contains JSON-encoded backslashes (binary PNG/PEM bytes from a real upload). Replaced with a single character-by-character logfmt unescaper that walks themsg="..."field byte-for-byte, finding the closing unescaped"and handling\\→\and\"→"once each. Tolerates trailing logfmt fields (vu=,iter=, etc.). (2) For multipart Content-Type, the k6 capture path now NEVER ships raw body bytes through JSON.stringify — even with the parser fixed, k6’s logfmt encoder can’t escape every binary byte JSON.stringify produces. Instead the captureExchange function walks the boundary chain inline (cap 100 parts) and emits a structured summary<multipart/form-data; boundary=...; N part(s); total M bytes: 'name':'filename' (Content-Type, N bytes), ...>derived from the actual parts. Strictly more useful than the raw envelope and always JSON-safe. Real-binary verified againstmockforge servewith a 2-file multipart upload:bodyfield now reads<multipart/form-data; boundary=...; 2 part(s); total 719 bytes: 'json':'json.json' (application/json, 18 bytes), 'photo':'photo.png' (image/png, 307 bytes)>. Four new regression tests pin the parser behaviour (logfmt round-trip with backslash escapes, raw mode, trailing-field tolerance, double-backslash adjacent to escaped quote). - [Contracts] Server-side violation
summarynow shows the actual + expected value inline instead of just the bare rule name (#79 round 46 / Srikanth on 0.3.190: “Here Not able to understand what is actual value in the request and what is expected”). Newinfer_got_expectedwalks the validator’sdetails[].messagetext and pulls the quoted got/expected pair, producingquery.lang (enum: got "zzz", expected one of "en", "fr")instead ofquery.lang (enum). Same logic mirrored into the TUI’s local copy so an older server still gets the upgraded summary at render time. Real-binary verified againstmockforge serve: a POST with 3 mixed query+body violations recordssummary: "3 query violation(s): body (enum: got \"basic\", expected one of \"free\", \"paid\"), query.$.xgafv (enum: got \"zzz\", expected one of \"1\", \"2\"), query.prettyPrint (type: got \"abc\", expected type \"boolean\")". - [Contracts] Classifier priority order now documents and handles HTTP-method violations (#79 round 46 / Srikanth on 0.3.190: “In the classifier order you mentioned: query > header > cookie > path > body — Where is http method violation coming?”). Method-not-allowed rejections (where the spec declares the path but not the method) emit a prose-only reason with no structured
"path"field, so the old per-loc chain missed them. Added an explicit check that runs AHEAD of the per-loc chain: full order is nowcontent-type > http-methods > query > header > cookie > path > body.
Added
- [Contracts] New
conformance-request-violations-by-request.jsonsidecar grouping every wire-level violation by(check, method, url)(#79 round 46 / Srikanth on 0.3.190: “I see three different messages, is this message for 3 different requests or for 1 request. if it is 1 request can we have 1 line item mentioning violation 1 = message1, violation2 = message2”). One row per emitted request; each row carries aviolation_countplusviolation_1,violation_2, … entries. The original flatconformance-request-violations.jsonstays as-is for tooling that wants the per-violation shape. Real-binary verified againstmockforge servewith a 2-op spec mixing query enum / query type / body enum: 30 flat entries collapse to 10 grouped requests (each carrying 3 violations). - [Contracts]
mockforge bench --auth-bearer <TOKEN>shortcut (#79 round 46 / Srikanth on 0.3.190: “If I have LDAP(tried user:password did not work), SAML, JWT Auth, How Should I send user credentials…”). Folded into the existing--conformance-headerplumbing so it flows through every conformance / self-test / multi-target path without separate wiring. De-dupes against an explicit--conformance-header "Authorization:..."if the user already supplied one. Help text documents the LDAP fallback (--conformance-basic-auth user:passfor LDAP gateways accepting Basic) and the SAML pattern (exchange the assertion for a session cookie at the IdP first, then forward via--conformance-header "Cookie: <session>"). Real-binary verified againstmockforge servewith--auth-bearer "<jwt>": every emitted request carriesAuthorization: Bearer <jwt>in its export. - [Contracts] New
conformance-network-events.jsonlog captures wire-level failures with timestamps and a classifiedkind(#79 round 46 / Srikanth on 0.3.190: “Is there a way or logs in mockforge client to understand timestamp range of traffic drops and client reason for traffic drop”). The native conformance executor’srequest.send().awaiterror path now pushes one entry per failure ({timestamp, check, method, url, kind, message}) and drains the sink to disk at the end of the run.kindis classified by reqwest’s typed errors intoconnect(TCP refused / no route),timeout(request didn’t complete in the configured budget),tls(cert / handshake rejection),body/decode/request, withotheras the catch-all. Real-binary verified by pointing the bench athttp://127.0.0.1:1(closed port): 10 wire-level events recorded, all classified askind: "connect", every entry stamped with an RFC-3339 timestamp.
[0.3.190] - 2026-06-23
Fixed
- [Contracts] Request/response body exports raised from 2000 bytes to 65536 bytes with a clear truncation marker (#79 round 45 / Srikanth on 0.3.189: “Now I see it is showing some details with one of the partial multi file upload data” with a 13MB rl_app_prof.json cut mid-PEM). The 2000-byte cap was inherited from r41 when bodies were typically small JSON; multipart uploads with even one PEM cert immediately overran it. All 8 captureExchange / captureFailure body capture sites (4 in
generator.rs, 4 inspec_driven.rs, both native executor capture sites inconformance/executor.rs) now cap at 65536 bytes and append<truncated at 65536 bytes; full body was N bytes>so the consumer can tell a truncation from a full payload. Real-binary verified againstmockforge servewith a 100KB upload: captured 65591 chars (65536 body + marker), marker reports the original 100246 bytes. - [Contracts] TUI conformance export now carries the
summaryfield (#79 round 45 / Srikanth on 0.3.189: “I am not seeing any summary info” in his exported TUI JSON). v0.3.189 addedsummaryto the server-sideServerConformanceViolationbut the TUI dedup wrapper (DedupedViolation) silently dropped it during the LIFO collapse. Now persisted on every dedup group — lifted verbatim from the server when it arrived populated (newer mockforge servers), or rebuilt locally fromreasonvia a TUI-residentsummarize_reason_localwhen an older server skipped the field (older mockforge). Three new unit tests pin all three branches. Also fixed a label mismatch where the summary saidrequest-body violation(s)while the violation’scategorysaidquery: both now follow the classifier priority order (query > header > cookie > path > body) so the labels agree. - [Contracts]
--validate-requestsnow actually fires when paired with the NATIVE executor (#79 round 45 / Srikanth on 0.3.189: “I am still not seeing any conformance failure logs or conformance-request logs are also not capturing any failures info”). Round 44 wiredvalidate_emitted_requestsinto the--use-k6branch only; Srikanth’s bench command doesn’t pass--use-k6, so his run never reached it. Now wired into all three execution paths: single-target k6, single-target native, multi-target native. Plus broadened to validate REQUEST BODY (top-levelrequired+ enum/type on direct string properties) and PATH PARAMETERS (extract from URL via template zipping) on top of round 44’s query params. Real-binary verified againstmockforge servewith a 2-operation spec mixing query enum, query type, path enum, and required body enum:--validate-requests --export-requests(no--use-k6) wrote 32 entries toconformance-request-violations.jsoncovering all 4 violation kinds. Also takes--base-pathso the Apigee-style/api/v1/...on the wire matches/v1/...in the spec.
[0.3.189] - 2026-06-22
Fixed
- [Contracts] k6 chain-context cookies / vars now persist across
chain_iterations(#79 round 44 / Srikanth on 0.3.188: “I see in 2nd and 3rd GET Request Cookie is going… below screenshots send empty cookie for all 3 GETs”). v0.3.186’s round-41 fresh-jar work hoisted thelet __ctx_cookie_X = ''declarations INSIDE thefor (let __iter ...)loop, so every iteration created a fresh empty binding that shadowed whatever the previous iteration had captured. Captures from iteration N were invisible to GET in iteration N+1, so a chain likeextract.cookies: [albsessid]on POST followed byCookie: "albsessid=${cookie:albsessid}"on GET emittedCookie: albsessid=on every GET. The declarations now sit at GROUP scope (above the iter loop) so a single binding persists across iterations. Real-binary verified against a Python mock that issues uniquealbsessidper request: GET 1 sendsalbsessid=(no capture yet, correct), GET 2 carries iteration-1’ssess-1-..., GET 3 carries iteration-2’ssess-2-..., all 48 POSTs carry the correct iteration-N capture. - [Contracts] Multipart upload exchange capture is now resilient to JSON.stringify failures (#79 round 44 / Srikanth on 0.3.188: “MOCKFORGE_UPLOAD_PARTS appears in console and k6-output.log but the request having this URL is sent out by k6 … is not getting captured neither in conformance-requests.json log nor conformance-failure-details.json”). For a multi-file upload with body bytes that JSON.stringify can’t roundtrip (lone surrogates, very long URLs, etc.), the primary stringify could throw and the entire
MOCKFORGE_EXCHANGEline was lost so the request disappeared from the export.__captureExchange(in bothgenerator.rsandspec_driven.rs) now wraps the build + stringify in try/catch and on failure emits a minimalMOCKFORGE_EXCHANGEentry with{check, method, url, status, _export_error}so the row never silently drops. A last-resort hand-rolled JSON literal covers the catastrophic case where even the fallback stringify fails. Applies to both the custom-checks generator and the spec-driven self-test generator.
Added
- [Contracts]
--validate-requestsnow also retrospectively validates each emitted request against the OpenAPI spec when paired with--export-requests(#79 round 44 / Srikanth on 0.3.188: “Any reason why validate-requests in mockforge client are not catching all this query param or body params or path params violation issues and record in conformance-request-failure logs?”). The existing pass only validates the custom YAML SHAPE at config time (missing path / required param / wrong method) — it never inspected the actual wire-level requests the bench fired. Newvalidate_emitted_requestsruns after the k6 process exits: walksconformance-requests.json, matches each entry to the spec, and emits aRequestViolationper query value that violates the parameter’s schema (enum / type:integer / type:number / type:boolean). Output is appended toconformance-request-violations.jsonso a single file holds both the YAML-shape pass and the wire-level pass. Real-binary verified against a 1-operation Apigee-style spec with alangenum: 13 emitted-request violations recorded across all self-test probes that sent?lang=test-value. - [Contracts] Server-side conformance violations now carry a short
summaryfield next to the fullreasonJSON (#79 round 44 / Srikanth on 0.3.188: “What is the difference between Validation error: and errors… both the errors are overwhelming to view”). The TUI / admin API / report tables can show the summary at a glance instead of unpacking the validator’sdetails+errorspayload. Shape:"<N> <category> violation(s): <name> (<rule>), ..."with<rule>mapped from common validator messages (enum,type,min,max,pattern,required,schema). Up to 5 items listed before truncating with+N more. Built once atrecord()time and persisted on theServerConformanceViolationstruct alongsidereason; older clients that read the buffer JSON see an emptysummaryand can fall back toreason.
[0.3.188] - 2026-06-21
Fixed
- [Contracts]
mockforge servenow emits response headers declared inresponses.<code>.headersinstead of silently dropping them (#79 round 43, surfaced during the round-42 bench-vs-serve sweep). A spec likeresponses.200.headers.Set-Cookie: { schema: { type: string } }would return200 OKfrommockforge servewith the cookie missing — the validator’svalidate_response_headersalready understood the shape but the synthesiser was never wired up, so any contract test or k6 chain that relied on the server issuing aSet-Cookiewould skip the cookie-based path entirely. NewOpenApiRoute::mock_response_headers_for_statuswalks the matched response’sheadersmap and produces schema-aware placeholder values: cookie-shaped (mockforge_session=mockforge-synthetic; Path=/) forSet-Cookie, the nil UUID forformat: uuidstrings, RFC-3339 epoch forformat: date-time,0/truefor integer / boolean, generic placeholder for plain strings, andmockforge-mock-valuefor any unresolved$ref. Both router builders (build_router_with_contextandbuild_router_with_mockai) merge the synthesised list into the response viainject_spec_response_headers, preserving axum’s existingcontent-type/vary/x-rate-limit-*headers (never overwrites). Three new unit tests pin the contract: cookie shape, UUID format, and the unmatched-status no-op. Real-binary verified:GET /loginwith the test spec returnsset-cookie: mockforge_session=mockforge-synthetic; Path=/andx-request-id: 00000000-0000-0000-0000-000000000000;POST /auth/login(MockAI router path) likewise emitsset-cookie.
[0.3.187] - 2026-06-21
Fixed
- [Contracts]
mockforge serveno longer rejects everymultipart/form-dataupload with400 body: Request body is required but not provided(#79 round 42 / Srikanth on 0.3.186, surfaced while re-verifying the round-41 multi-file upload report againstmockforge serveinstead of a hand-rolled mock). The MockAI-enabled OpenAPI router (build_router_with_mockai, the active path since v0.3.176) only ranserde_json::from_sliceon the raw body, so a multipart envelope quietly failed to parse,bodystayedNone, and the spec’srequestBody.required: truemade the validator emit “body required” before the request ever reached the multipart extractor. The router now sniffsContent-Typeformultipart/form-dataand runsextract_multipart_from_bytesto build a synthetic JSON object (one entry per form field) so the validator sees the body exists and schema validation can run against the fields. File parts contribute their tmpfile path. Empty / malformed multipart envelopes still surface as “body present” so the validator does not 400 — schema mismatches surface as their own validation errors downstream. Real-binary verified againstmockforge serve --spec uploads.json --http-port Nwith 9 distinct file parts (json / html / css / js / xml / csv / txt / md / pdf): every part hits the server and the request returns200 OKinstead of the prior400. Without this fix the round-41 multi-file bench cannot exercisemockforge serveend-to-end — it 400s before any check runs.
[0.3.186] - 2026-06-21
Fixed
- [Contracts] k6 custom checks now use a fresh per-request cookie jar so
${cookie:NAME}substitution is the only source of truth (#79 round 41 / Srikanth on 0.3.185: “I see there are two cookies getting sent in all the 16 POST Requests”). v0.3.184 wired the substitution but k6’s per-VU jar ALSO auto-collected the GET response’sSet-Cookieand re-injected it on every subsequent request, so the POST went out with two copies ofalbsessidin itsCookie:header. Now a__custom_jar_factory = () => new http.CookieJar()lives at init scope and every chain request carriesjar: __custom_jar_factory()in its params. Real-binary verified: 48 POSTs across 3 chain iterations land with a singleCookie: albsessid=abc-xyzeach. Also addresses his “have … without Cookie in GET” ask: with no shared jar, iteration K+1 GET does not carry cookies from iteration K unless the user explicitly forwards them via${cookie:NAME}. - [Contracts] Chain-iterations summary aggregation now sums per-occurrence executions instead of unique check names (#79 round 41 / Srikanth on 0.3.185: “Configured Chain Iteration = 3 but I see requests are getting sent 3 times in conformance-request logs but in the category column I see it shows 2”). With
chain_iterations: 3andcustom:get+custom:post (repeat 16 parallel), the per-request log records 51 requests but the summary table previously printedCustom: 2 / 0 / 2 / 100%because each unique top-level check name was added with+1. Now a newis_primary_custom_checkhelper skips:header:/:body:sub-checks (so a check with N header assertions does not appear as N+1 requests) and the row sums(passes, fails)per primary check. Same YAML now printsCustom: 51 / 0 / 51 / 100%— exactly the 3 GET + 48 POST that went on the wire. - [Contracts] Validator violation category labels now reflect WHERE the failure happened, not the prose of the error message (#79 round 41 / Srikanth on 0.3.185: “GET (ideally should not have request body so GET falling under request body category is little weird [sic]) and POST (ideally should not have query params but I see violation on query params)”). The classifier in
mockforge-openapi::openapi_routes::classify_validation_reasonnow reads the structured"path":"<loc>.<name>"field from the validator’s JSON details FIRST, so a query violation getscategory: query, a header violation getsheaders, a cookie one getscookies, and only body violations stayrequest-body. The older heuristic remains as a fallback for callers that do not embed a structured path. Real-binary verified against the Google Apigee spec from #888:POST /v1/organizations?$.xgafv=test-valuerecordscategory: query(wasrequest-body). - [Contracts] k6 multi-file upload now logs every part it is about to send +
--export-requestsshows a multipart summary instead of an empty body (#79 round 41 / Srikanth on 0.3.185: “I could only see 5 out of 9 Files in the PCAP. Is there a way from Logs I can confirm all the files in multiple upload are sent from mockforge client” + “When run without Spec the export request file has blank entry”). Each upload check now emits one tagged log line per check at request-build time (MOCKFORGE_UPLOAD_PARTS: <name> <N> files: <field>:<filename> (<ct>), ...) so agrep MOCKFORGE_UPLOAD_PARTSin k6 stdout confirms every part left the client.--export-requestsnow surfaces a brief<multipart upload; body bytes not surfaced by k6 res.request.body>string when k6’sres.request.bodyis empty (it is for multipart uploads — k6 serialises the form internally and does not expose the bytes JS-side), so the export no longer shows a blankrequest.bodyfor multi-file checks.
[0.3.185] - 2026-06-20
Fixed
- [Contracts] OpenAPI request validator now picks up path-level parameters and resolves
$refparameters, so query / header / cookie violations are detected on specs that follow the OpenAPI 3.0 §4.7.10.1 inheritance pattern (#888 / #79 round 40 / Srikanth on 0.3.183: “Our Proxy is detecting below Violation … But mockforge server and client doesnt show this violation”). The Google Apigee spec he attached defines the shared auth / format query params ($.xgafv,prettyPrint,key,oauth_token, etc.) once at PATH level via$refto#/components/parameters/*, then inherits them on every operation. v0.3.184’sspec::operations_for_pathcloned eachOperationbut ignoredpath_item.parameters, so the inherited params never reached the validator; even when they did, the validator’sp_ref.as_item()call skippedReferenceOr::Referenceentries, so a path-level enum violation walked past validation silently. Two helpers inmockforge-openapi::spec:merge_path_params_into_operation(operation-wins-on-collision per the spec) andresolve_parameter_ref(one-level$refwalk with tail-resolve for chained refs). Both run once at registry build time insideoperations_for_path, so both registry builders (openapi_routes.rsandopenapi_routes/registry.rs) pick up the merged + resolved list. New unit test pins the override semantics against a minimal repro spec. Real-binary verified against Srikanth’s exact spec:POST /v1/organizations?$.xgafv=test-value&prettyPrint=test-valuewent from200 OK(silently passed) to400with both violations recorded on/__mockforge/api/conformance/violations('$.xgafv' validation failed: "test-value" is not one of "1" or "2",'prettyPrint' validation failed: "test-value" is not of type "boolean").
[0.3.184] - 2026-06-20
Fixed
- [Contracts]
mockforge bench --use-k6now emits a working multipart upload when a custom check hasupload/uploads(#79 round 39 / Srikanth on 0.3.183: “I tried with single file upload … I see file data is not getting upload”). v0.3.183 only wired the YAML’s file-upload knobs into the native executor, so the same YAML under--use-k6produced a k6 script that calledhttp.post(url, null, {})and the file content was lost entirely. The k6 generator now hoistsconst __file_N = open('...', 'b')to script-init scope (k6 rejectsopen()inside the VU function) and folds everyhttp.file()part into a singlehttp.post(url, form, { headers })call. Mid-config warning surfaces (bodywins overupload/uploads) are emitted as JS comments next to the request line. - [Contracts]
mockforge bench --use-k6now honoursextractcapture and${var:NAME}/${cookie:NAME}/${header:NAME}substitution end-to-end (#79 round 39 / Srikanth on 0.3.183: “I am seeing some unknown data in Cookie” / “I am not seeing 16 Request of POST being sent in parallel rather seeing only 1 Request”). Chain context variables (let __ctx_cookie_X = '',let __ctx_var_X = '') are pre-declared at the top of the group so a later check that reads a captured value never throwsReferenceError. After each check’slet res = ...line the emitter writes__ctx_cookie_X = res.cookies['X'][0].valuefor cookies, a case-insensitive header lookup for__ctx_var_X, and a parsed-JSON dotted-path walk for body fields. Substitution itself produces template-literal-safe text in one pass —${cookie:session}becomes${__ctx_cookie_session}(intentional interpolation), unrelated${...}shapes escape to\${...}so they end up as literal text in the request without breaking out of the template.repeat: count: N, mode: parallelnow emitshttp.batch([N entries])(k6’s natural concurrency primitive) so 16 parallel writes fire as one batch instead of one request;mode: sequentialemits aforloop. Both branches honour multipart uploads and chain-context substitution. Real-binary verified against a local mock: 16 POSTs land carrying the capturedalbsessidcookie, body is the expected JSON, body length matches. - [Contracts] Reference checks (the 47 built-in
/conformance/...probes) are skipped when the user passes--conformance-customwithout--specand without an explicit--conformance-category(#79 round 39 / Srikanth on 0.3.183: “In the exported request I see it is sending request to api/conformance/params/hello and some other URLs”). v0.3.183 ran the customs ALONGSIDE the references against any target, so the request log filled with irrelevant 404s on/api/conformance/params/helloetc. The skip applies to both the native executor (NativeConformanceExecutor::execute) and the k6 generator (ConformanceGenerator::generate) so the behaviour matches across execution modes.
[0.3.183] - 2026-06-19
Added
- [Cloud]
mockforge serve --admin-host <addr>lets the admin port bind on a non-loopback interface so amockforge-tui(or admin UI in a browser) on another machine can connect (#79 round 38 / Srikanth on 0.3.182: “admin ports are opening only on 127.0.0.1 so I am able to open the mockforge tui UI on the Server locally but when I access the server from some other machine via mockforge TUI connections are getting refused”). Default stays loopback for safety. Pass0.0.0.0to expose on every IPv4 interface or::to dual-stack bind on IPv6 + IPv4-mapped-IPv6. Real-binary verified:--admin-host 0.0.0.0binds the admin port on0.0.0.0:N, andcurlfrom the host’s eth0 IP returns 200 on/__mockforge/api/conformance/violations. - [Contracts] Bench custom checks now support file uploads as
multipart/form-data(#79 round 38 / Srikanth on 0.3.182: “Do we have a way to do Single file upload and multi file upload options(All types of content-type or file extensions example .exe, .jpg, .json, .docx, .xml etc) in mockforge bench”). New YAML fields onCustomCheck:upload: { path, content_type, field_name, filename? }for a single file,uploads: [{...}, {...}]for multi-file requests. Bytes are read off disk at construction time so a missing file is surfaced before the run starts (not mid-stream). NewCheckBody::Multipartvariant +reqwest::multipart::Formwiring in the executor’s send path;mime_strfailures fall back toapplication/octet-streamso a typo incontent_typestill sends bytes. Brief summary in--export-requestsoutput (<2 file(s): a.jpg (image/jpeg, 1234 bytes), b.json (application/json, 56 bytes)>) instead of dumping raw multipart envelopes.bodywins overupload/uploadswhen both are set (mistake gets a stderr warning). - [Contracts] Bench custom checks now support cookie + CSRF capture and reuse across a chain of requests (#79 round 38 / Srikanth on 0.3.182: “Get a cookie and csrf token on first request from the server, use that cookie and csrf token in 16 subsequent requests that should be sent in parallel” Sequence 1 + sequential Sequence 2, iterable any number of times). New YAML fields on
CustomCheck:extract: { cookies: [...], headers: { var: header }, body_fields: { var: dotted.path } }captures values from the response;repeat: { count: N, mode: parallel | sequential }fires the check N times (parallel usestokio::join_all; sequential runs one-after-another). Top-levelchain_iterations: NonCustomConformanceConfigrepeats the wholecustom_checkslist N times, with a fresh chain context per iteration (no leakage between iterations). Substitution tokens in path / headers / bodies:${var:NAME},${cookie:NAME},${header:NAME}. Unknown tokens preserved verbatim so a missing capture is visible in the request log rather than silently sending an empty string. JSON body walks only touch string leaves so numbers / arrays / booleans are unchanged. Multipart bodies pass through unchanged. 4 new unit tests cover substitution and extraction (token shapes, missing-capture graceful skip, fullConformanceCheckround-trip). Real-binary verified end-to-end: 2-iteration chain (login + 4 parallel + 3 sequential do-work) produces login: 2, do-work-parallel: 8, do-work-sequential: 6, total: 110;${var:user_id}was substituted to42from the login JSON response,${var:csrf}left verbatim because the mock server doesn’t synthesizeX-CSRF-Token(correct: missing capture stays visible).
Documentation
- [Contracts] New
book/src/reference/bench-custom-checks.mdpage with the YAML reference for the round-38 features (file uploads, cookie/CSRF chain, repeat semantics, chain iterations) plus copy-paste recipes for Srikanth’s exact Sequence 1 (parallel 16 writes after login) and Sequence 2 (sequential writes after login) scenarios. - [Cloud] New “Accessing the admin port from another machine” subsection in
book/src/getting-started/quick-start.mddocumenting--admin-host 0.0.0.0+ the security note about pairing it withauth_required: trueoutside trusted networks.
[0.3.182] - 2026-06-18
Added
- [DevX] TUI conformance-violation exports now carry the bench client’s
mockforge_version+client_sent_atso a user grepping the export can line a row up with the bench JSONL row that produced it (#79 round 37 / #876 follow-up / Srikanth on 0.3.181: “client time is 24 hour format and mockforge tui exported logs has time 12 hour format”). v0.3.181 added the two fields to the server-sideServerConformanceViolation, but the TUI’sDedupedViolationdropped them on export, so the only timestamps in the file were the server’sfirst_seen/last_seen(when the server recorded the violation), not the client’sclient_sent_at(when the bench sent the probe). Three changes: (1)ConformanceViolation(TUI’s mirror struct) deserializes the two new server fields; (2)DedupedViolationgainsclient_mockforge_version+min_client_sent_at/max_client_sent_atOption fields (skip_serializing_if for back-compat with older servers); (3) dedup logic extracted to a purededup_violationshelper so the aggregation can be unit-tested. The min/max bracket the client-side send window across all hits in the dedup group; the version is the latest observed (handles upgrade-mid-run). Three new unit tests cover happy-path aggregation, back-compat with unstamped servers, and mixed-version groups. - [DevX]
mockforge-tui --admin-url URL1 --admin-url URL2now adds all URLs to a server rotation (#79 round 37 / Srikanth on 0.3.181: “Can I connect Mockforge tui to multiple Server IPs and see Logs at the same time by switching Tabs based on Server IPs”). Inside the TUI,Ctrl-]cycles to the next server,Ctrl-[to the previous, and the header shows[2/3] http://h2:9080so the user always knows which server’s data they are looking at. Single-server runs (the default) are unchanged: the rotation key is a no-op and the header indicator falls back to the original single-URL layout.TuiConfiggainsextra_servers: Vec<String>(#[serde(default)]) +all_admin_urls()for dedup + primary-first ordering;App::rotate_server(step)swaps the activeMockForgeClientand backdateslast_health_checkso the connectivity indicator updates on the next tick.keybindings::Action::{NextServer, PrevServer}map toCtrl-]/Ctrl-[; help overlay documents both. Four new tests cover the rotation: wrap forward + backward, single-server no-op, and the dedup/primary-first ordering.
Documentation
- [DevX] New “Recipes” section in
book/src/user-guide/chaos-engineering.mdwith copy-paste recipes for mixed 4xx/5xx via--chaos-http-errors, slow backend via--chaos-latency-ms/--chaos-latency-range, real timeouts (504 after a sleep) via thetimeout_errorsYAML knob, per-endpoint chaos viaroutes-with-chaos.yaml(multiplefault_typesentries pick uniformly so a list of three statuses gives ~equal weight), and a retry/backoff exercise that combines them (#79 round 37 / Srikanth Q-c: “Can you share some examples where for some of the APIs or API Endpoint(Method + Path) in specs we have to send different 5xx responses from Server using mockforge serve –chaos and Also slow responses from Server so that Client Timeouts for few APIs or API Endpoints”). Plus a note about confirming what actually fired viaGET /api/chaos/status. - [DevX] Windows install path documented in
book/src/getting-started/installation.md(#79 round 37 / Srikanth Q-d carry-over: “Do we have mockforge-tui installer for Windows?”). First-class signed Windows binaries are not published today; the working install path iscargo install mockforge-cli --lockedwith Rustup, or Docker Desktop / WSL2 for users who prefer no compile step. First-class signed Windows binaries (no compile step, no SmartScreen warning, winget) tracked at #884.
[0.3.181] - 2026-06-17
Fixed
- [DevX] Capture-viewer HTML (
conformance-self-test-requests.html) status badge colour now agrees with theexp 2xx-4xxbadge it sits next to (#79 round 36 / #875 / Srikanth on 0.3.180). Round 35 made variant-b embedded-content probes pass on 4xx in the JSONL + JSON report, but the per-card badge in the capture viewer still rendered a 400 in red becausestatusClass(c)colored by the raw 200..400 range, ignoringexpected_status_range. NowstatusClassconsultsisMismatch(c)whenever the capture knows its expected range (older captures without the field fall back to the round-23 raw-range behaviour for backwards compat). The top-of-page summary line and the “matched expected” / “mismatch” filter checkboxes use the same logic via a newprobe_passedRust helper. Two new unit tests cover the2xx-4xx/4xx/2xx-3xx/ legacy / error branches ofprobe_passed, and snapshot the summary line for a mixed variant-b + positive capture. Real-binary verified against Srikanth’s exactPOST /api/appliance/local-accountsscenario.
Added
- [Contracts] Conformance probes now stamp the client mockforge version and send-time on both sides for cross-correlation (#79 round 36 follow-up / #876 / Srikanth’s r36 ask: “good to also have what mockforge tui version the request came from and at what time the request was sent … on both client and server logs”).
mockforge bench --conformance-self-testsets two outbound headers on every probe:X-Mockforge-Client-Version(fromCARGO_PKG_VERSION) andX-Mockforge-Client-Sent-At(RFC3339). The same values are written into the JSONL line asmockforge_version+client_sent_at(both#[serde(default)]so older captures still load) and rendered as a dedicatedClient:row in the capture HTML viewer. On the server side,ServerConformanceViolationgainsclient_mockforge_version: Option<String>+client_sent_at: Option<DateTime<Utc>>, populated by a sharedread_client_stampshelper at every header-bearing record site (content-type mismatch + body validation on both router builders). The admin endpoint/__mockforge/api/conformance/violationssurfaces both fields. Header constants live inmockforge-foundationso bench and openapi cannot drift. Real-binary verified end-to-end: a probe sent atTshows the samemockforge_versionandclient_sent_atin the bench JSONL row, in the request headers, in the admin endpoint response, and in the capture HTML.
Documentation
- [DevX] New
book/src/reference/conformance-self-test-probes.mdsections covering the content-type-mismatch probe family (rounds 27 / 28 / 32,request-body:content-type-mismatch:{xml,yaml,multipart,urlencoded}, expected415) and the variant-b embedded-content family (round 27 / 35,request-body:embedded-content:*, expected2xx-4xx, only5xxis a finding). Rewrote the “How ‘passed’ is decided” section to describe the tristateExpectedOutcomeenum (Success/ClientError/NotServerError) introduced in round 35, with the per-probe assignment table. New “What--inject-response-violationschanges” section clarifies that the flag modifies response bodies only (drops the first declared required field on synthesized 2xx responses) and does NOT change response status; points users atmockforge serve --chaosandmockforge route-chaosfor status-class chaos. Notes that the bench never deliberately elicits a5xx, so a5xxis always a server-side finding (#79 round 36 follow-up / #877 / Srikanth’s r36 ask for a published probe taxonomy).
[0.3.180] - 2026-06-16
Fixed
- [Contracts] Embedded-content variant-b probes (
request-body:embedded-content:xml/yaml/multipart/urlencoded) no longer flag 4xx responses as mismatches (#79 round 35 / #859 / Srikanth on 0.3.179: 4 noise rows onPOST /api/appliance/local-accountsand ~268 similar rows across the vCenter spec). The round-27 design only ever wanted to catch 5xx server crashes; a 4xx from the server’s pattern / format validator on the field we inject into is correct behavior. Replaced the olderexpected_4xx: boolflag onsend_case/send_case_with_extrawith a tristateExpectedOutcomeenum (Success/ClientError/NotServerError); variant-b probes opt intoNotServerErrorso the pass calculation isactual_status >= 200 && actual_status < 500.expected_status_rangeonCaseCapturecarries the new"2xx-4xx"value, which the HTML viewer +jqfilters pick up automatically.CaseOutcome.expected_4xxkeeps its old semantics (negative-probe flag), soreport_html.rsis unchanged. Reproduced locally against the exactPOST /appliance/local-accountsroute Srikanth flagged: 4 embedded-content rows go from 4 misses to 4 passes on v0.3.180. Two new unit tests cover the pass-rule and the string-label table.
[0.3.179] - 2026-06-15
Fixed
- [DevX]
response_schema_errorstripsdescription/example/examples/summary/title/externalDocs/xmlfrom the focused schema before serializing, so the actually-useful constraint keywords (type,required,properties,format,min*/max*,pattern,oneOf/anyOf/allOf/not) survive the 300-char truncation cap (#79 round 34 / #827 / Srikanth on 0.3.178: vCenter’senabled: booleanproperty had a multi-paragraph description that ate the budget before"type":"boolean"could appear). Two new unit tests assert (1) the vCenter-shaped case keeps"type":"boolean"in the printed schema and (2)strip_schema_noisekeeps every constraint keyword while dropping every prose key. - [Contracts] Per-endpoint summary
pathcolumn now matches the URL the user actually sent (#79 round 34 / #828 / Srikanth on 0.3.178: searched for/api/appliance/access/consolecliinconformance-per-endpoint.jsonand didn’t find it because round 33 stored just the spec template/appliance/access/consolecli). The bench now constructspath_templateasbase_path + op.pathand stamps that on everyCaseCapture, matching the same prefix logicbuild_url_with_basealready uses. New unit test asserts the prefixed value flows through toPerEndpointSummary.path. - [Contracts] Embedded-content variant-b probes are skipped when the positive sample body has no string field (#79 round 34 / #829 / Srikanth on 0.3.178: PUT
/api/appliance/access/consolecliexpects{enabled: boolean}but the round-27 fallback envelope{"data": <snippet>}was structurally different, so the server correctly 400’d and the bench misreported the2xx-3xxexpectation as a miss).embed_payload_in_first_string_fieldnow returnsOption<String>; the variant-b loopcontinues when the helper returnsNone. Three test cases cover Srikanth’s exact{"enabled":true}shape, the no-string scalar-only case, and invalid-JSON samples.
[0.3.178] - 2026-06-14
Added
- [Contracts] Per-endpoint summary now groups by the spec’s path TEMPLATE rather than the resolved URL, and rows carry a
speclabel for multi-spec runs (#79 round 33 / #823 / Srikanth’s r32 follow-up). On a v1 run that exercised/users/Xand/users/Yas 1000 distinct VU iterations, the report exploded to 1000 rows; on v2 those collapse into a single(GET, /users/{id})row.CaseCapturegainspath_template: Stringandspec_label: Option<String>(both default-via-serde so older JSONL captures still load). HTML section only renders the Spec column when at least one row has a label, so single-spec runs stay tidy. New tests cover template collapsing, multi-spec separation, the HTML column toggle, and legacy-fallback to resolved path. - [Contracts]
MOCKFORGE_INJECT_RESPONSE_VIOLATIONS=trueenv var +mockforge serve --inject-response-violationsCLI flag (#79 round 33 / #822 / Srikanth’s r32 follow-up: “can we add those as a negative response tests from mockforge server side”). When enabled, the OpenAPI response generator drops the first declared required field from every synthesized 2xx response body, so the caller can exercise their proxy / conformance pipeline against a known-bad-shape mockforge end-to-end. Off by default. No-op for non-2xx, non-Object schemas, schemas withoutrequired, and bodies that already don’t carry the field.
[0.3.177] - 2026-06-14
Fixed
- [Contracts] Server-side conformance buffer now records
content-typesviolations from the MockAI-enabled router path too (#79 round 32 / Srikanth on 0.3.176: “in the client I see response status as 415 for labelrequest-body:content-type-mismatch:xmlrequest but in the server I see only 400”). Root cause: the MockAI request handler tookbody: Option<Json<Value>>, and axum’sJsonextractor 415s a request whoseContent-Typeisn’tapplication/jsonBEFORE the handler runs, so the buffer never got a chance to record the violation. The handler now takes rawaxum::body::Bytes, runs the samecheck_request_content_typeprecheck the non-MockAI router already had (records the violation with categorycontent-typesand the configured validation status, default 415), and parses the body as JSON manually for the validator / fingerprint / MockAI paths. The existingcheck_request_content_type_flags_mismatchtest covers the logic; the wiring is inbuild_router_with_mockai. - [Install]
cargo install mockforge-clino longer requiresprotobuf-compileron the build host (#79 round 32 / Srikanth’s VM1 Ubuntu 16.04 install error “Could not findprotoc”).mockforge-grpc/build.rsnow picks theprotocbinary from theprotoc-bin-vendoredcrate when the user hasn’t pointedPROTOCat their own copy; user-setPROTOCstill wins (empty string treated as unset). Build verified against the vCenter spec build withPROTOCcleared from the environment.
Added
- [Contracts] Per-endpoint traffic summary derived from the conformance self-test capture (#79 round 32 / Srikanth’s Q6 ask). Two surfaces:
- JSON sidecar at
bench-results/conformance-per-endpoint.jsonfor automation pipelines. One entry per(method, resolved URL path)withsent,status_2xx/status_3xx/status_4xx/status_5xx/errors, plusrequest_body_len/response_body_len/query_lenwith samples / avg / p50 / p95 / max. - HTML section “Per-endpoint traffic summary” spliced into
bench-results/conformance-report.html. - Grouping is by RESOLVED path for v1, not the spec template (so
/foo/Xand/foo/Yare distinct rows). A future round will collapse to the spec template once we surfaceop.pathon eachCaseCapture. 4 new unit tests cover the bucketing, query-string strip, p95 calculation, and empty-input HTML.
- JSON sidecar at
[0.3.176] - 2026-06-10
Fixed
- [Contracts]
mockforge servePOST / PUT / PATCH / DELETE responses now match the spec’s response schema instead of MockAI’s hardcoded{id, status, data}envelope (#79 round 31 follow-up / Srikanth’s vCenterArchive.Infocommentfinding). Root cause: MockAI is wired on by default through the reality engine’sModerateRealismlevel (reality.enabled = true→mockai.enabled = true), and itsgenerate_response_bodyfor the Create / Update / PartialUpdate / Delete mutation kinds returned a hardcoded envelope that ignored the OpenAPI 2xx response schema. Result: every write operation came back as{"id":"generated_id","status":"created","data":<echoed body>}, missing whichever required fields the spec actually defined (e.g. the six fields on vCenter’sAppliance.Recovery.Backup.SystemName.Archive.Info). The function now returns{}for every mutation kind, which the calling site already handles as “fall through to the OpenAPI ResponseGenerator”, so the response body becomes spec-shape automatically and required fields are populated. New regression test enumerates everyMutationTypeand asserts each returns{}. Real-binary verified against the exact vCenter route from Srikanth’s r31 report: the response now containscomment,location,parts,system_name,timestamp,version.
[0.3.175] - 2026-06-10
Fixed
- [DevX]
response_schema_errornow shows the missing property’s own sub-schema forrequired field missingerrors instead of the entire parent object schema (#79 round 31 / Srikanth on 0.3.174 against the vCenterAppliance.Recovery.Backup.SystemName.Archive.Infospec). The prior output dumped the parent’s multi-paragraph description and every sibling property schema then truncated at 300 chars, which buried which field was missing and why. The walker now strips the surrounding quotes fromproperty(jsonschema’s Display impl wraps the name in"..."), descends one more step intoproperties[<missing>], and prints just that. New test covers the vCenter scenario and asserts the parent description + sibling names don’t leak into the suffix. - [Install]
cargo install mockforge-clino longer requires a system OpenSSL 1.1+ at runtime, so it works on Ubuntu 16.04 / RHEL 7 / boxes that only ship OpenSSL 1.0.2 (#79 round 31 / Srikanth’s “few clients which runs ubuntu 16.04 and openssl version 1.0.2”). Two fixes: (1) the TLS stack is now rustls-only (workspacereqwest,lettre, and thesqlxuses inmockforge-tunnel+mockforge-vbrall dropped*-native-tlsfeatures);cargo tree -i native-tlsagainstmockforge-clicomes back empty. (2)git2(viamockforge-plugin-loaderandmockforge-scenarios) was still pullingopenssl-sysfor its HTTPS clone path, so we enabledvendored-opensslon it; that statically links a known-good OpenSSL intolibgit2-sys/libssh2-sysso the binary no longer needslibssl.soat runtime. Also converted four crates that hand-wrotereqwest = { version = "0.12", features = ["json"] }to inherit the workspace dep so the default-tls drift can’t reappear.
Added
- [Server]
mockforge serve --conformance-buffer-size Nand--conformance-buffer-uniqueCLI flags (#79 round 31 / Srikanth: “is it possible give in the mockforge server command as opposed to environmental variable which I sometime forget”). Both mirror the round-29/round-30 env vars; flag values win when both are set. Surfaced inmockforge serve --helpunder “Server Configuration”.
[0.3.174] - 2026-06-09
Fixed
- [DevX]
response_schema_errornow prints only the sub-schema at the offending JSON Pointer instead of the full top-level schema (#79 round 30 / Srikanth on 0.3.173 asking what the message reads as for nested fields). A{"name": 123}mismatch against{"type":"object","properties":{"name":{"type":"string"}}}now readsresponse body at /name: expected type string; expected schema {"type":"string"}, not... expected schema {"properties":{"name":{"type":"string"}},"type":"object"}. Walker descends throughproperties(anditemsfor arrays) following the instance pointer; falls back to the full schema when the path can’t be resolved (additionalProperties, oneOf/allOf, unresolved $refs).
Added
- [Contracts]
MOCKFORGE_CONFORMANCE_BUFFER_UNIQUE=trueswitches the server-side conformance violation buffer from FIFO to dedup-by-signature (#79 round 30 / Srikanth on 0.3.173: “Can we have this buffer for unique violation as opposed to duplicate violation”). Every duplicate of an already-buffered(method, path, status, category, reason)hits its existing entry and bumps the newoccurrencesfield onServerConformanceViolationinstead of consuming a new slot — so 10M requests with 150 distinct violation kinds keep all 150 visible in a 256-slot buffer, instead of being clobbered by the most common offender. Newoccurrences: u32field on the violation struct (defaults to1on older payloads via serde default). TUI’s export-time dedup grouping now sumsoccurrencesper group so the headline counts reflect the true server-side hit count under unique mode. Three new tests cover dedup, multi-signature distinction, and FIFO-order eviction inside the unique buffer.
[0.3.173] - 2026-06-08
Fixed
- [DevX]
response_schema_errormessage now explicitly names the response body as the location (#79 round 29 / Srikanth on 0.3.172). The priorat /: expected type string; expected schema {...}read as a URL path to readers unfamiliar with JSON Pointer syntax; Srikanth couldn’t tell whether/referred to the URL root or the response body root. New format:response body root: ...for top-level andresponse body at /name: ...for nested fields. Two new tests cover the root and nested cases.
Added
- [Contracts]
MOCKFORGE_CONFORMANCE_BUFFER_SIZEenv var raises the server-side conformance violation ring buffer cap above the default 256 (#79 round 29 / Srikanth: TUI shows 10,145 violations seen but export only had 114 unique entries). Capped at 64k entries to keep peak memory bounded. Empty / zero / unparsable values fall back to the default. Includes an ignored-by-default unit test (effective_buffer_size_respects_env_var) that exercises the parser — run withcargo test -p mockforge-foundation -- --ignored --test-threads=1because it touches process-wide env. - [Contracts] Capacity advisory printed at bench start when
targets × VUs ≥ 150(#79 round 29 / Srikanth’s “5 VUs against 50 targets hung my 15 GB VM”). Printstargets,VUs, estimatedRPS-total, estimated CPU cores and RAM GB, and points to the new sizing doc. Heuristic, errs on the warning side; doesn’t block the run. - [DevX] New book page
reference/bench-capacity-sizing.mdwith a sizing table (1 → 100 targets, 10 → 500 RPS each), the per-VU / per-target RAM and CPU formulas, symptoms of under-provisioning, and ajq-based sharding recipe for >50 targets. Surfaced inSUMMARY.mdnext to the probe-label reference.
[0.3.172] - 2026-06-07
Fixed
- [Contracts] Content-type-swap probes now actually send only the wrong Content-Type (#79 round 28). My round-25 fix relied on reqwest’s
.header(k, v)being last-write-wins, but it actually APPENDS — so each content-type-swap probe was sending BOTHContent-Type: application/json(from the body block) ANDContent-Type: application/xml(fromextra_headers). Axum’sJson<>extractor picked the JSON one and accepted, server returned 204, the spec violation Srikanth was investigating never showed up in the conformance buffer. Nowsend_case_with_extrabuilds aHeaderMapourselves, where.insert()replaces, so the override actually replaces. Smoke-confirmed: the same 4 content-type-mismatch probes that returned 204 on 0.3.171 now return 415 on this build. - [Contracts] Server-side
OpenApiRouteRegistry::check_request_content_typeflags Content-Type mismatches against the spec’s declaredrequestBody.contentkeys (#79 round 28 / Srikanth’s main r27 ask). Called from the live-server route handler before the body schema validator; on mismatch records acontent-typescategory violation in the conformance buffer with the configured validation status (defaults to 415). Stripping the; charset=...suffix so the comparison is type/subtype only; case-insensitive matching. - [DevX]
response_schema_errornow embeds the expected schema in the message (#79 round 28 / Srikanth: “what does at /: mean? would be good to put asexpected schema {...}”). Format:at <path>: <kind>; expected schema <compact JSON>. Schemas are truncated to 300 chars to keep JSONL lines bounded.
Added
- [Contracts]
expected_status_rangefield on everyCaseCapture(#79 round 28 / Srikanth: “Is it possible to put expected response code status in both jsonl and jsonl report”). Values:"2xx-3xx"for positive probes,"4xx"for negatives. Persisted in the JSONL sojqcan filter mismatches, and rendered as a smallexp 4xx/exp 2xx-3xxbadge on every card in the HTML viewer. - [DevX] HTML capture viewer: new “only show mismatches” checkbox in the toolbar (#79 round 28 / Srikanth: “Also a filter or checkbox to get the request whose response code is not matching”). Filters across the whole capture (cross-page), composes with the status checkboxes and search box, recomputes pagination immediately.
- [DevX] HTML report: per-category counts in the per-operation
By categorycolumn are now clickable links (#79 round 28 / Srikanth: “Is it possible to give link to the count for By category column also”). Eachcat:Njumps to#miss-cat-<cat>in the drill-down table. Cap-aware: only emits links when the category has a surviving row in the truncated drill-down.
[0.3.171] - 2026-06-06
Fixed
- [DevX] Capture HTML viewer now paginates with cross-page filters (#79 round 27 / Srikanth d3). The round-25 + round-26 1000-card cap silently hid 4xx/5xx probes past the cap; he had to fall back to
jqto find errors. The viewer now embeds the full capture as a JSON array in a<script id="captureData">element, filters in JS over the whole array, and renders only the current page (50 cards) of the filtered subset. First/Prev/Next/Last buttons plus a “Jump to page” input. “(N of M shown)” banner appears when filters narrow the set. The 1000-card cap is gone; the JSONL still has the full set for grep / jq workflows.
Added
- [Contracts] Content-type swap variant (b) probes (#79 round 27 / Srikanth k variant b). Companion to the round-25 variant-a probes. Where variant (a) lies about Content-Type (sends XML/YAML/multipart/urlencoded body as
Content-Type: application/json), variant (b) keeps Content-Type honest asapplication/jsonand the body parses as valid JSON, but a string field’s value carries an XML / YAML / multipart / urlencoded snippet. Labels:request-body:embedded-content:xml/yaml/multipart/urlencoded. Expected status is 2xx-3xx because the envelope IS spec-shape; a 5xx flags a server that crashed trying to parse the embedded snippet (a real bug), while a 4xx flags a narrower-than-expected string field validator (often correct server behaviour). Helperembed_payload_in_first_string_fieldwalks the sample body depth-first, replaces the first string leaf with the snippet, and falls back to{"data": "<snippet>"}when no string field exists.
[0.3.170] - 2026-06-06
Fixed
- [Contracts] TUI Conformance tab: detail modal now snapshots the violation’s text at Enter time and renders from the snapshot, not from a live
selected_violation()lookup (#79 round 26 / Srikanth on 0.3.169 “still see same behavior”). My round-25 identity-key re-anchor only worked when the user’s clicked violation was still in the new buffer; under heavy traffic the 256-cap evicted it, leavingTableState.selectedat the same numeric index pointing at a different request. The snapshot captures the detail string when Enter fires (detail_snapshot: Option<String>) so the modal text is frozen at that moment; Esc drops the snapshot for the next Enter. Two regression tests cover the index-shift scenario AND the on_data refresh path. - [DevX]
response_schema_errorno longer renders Rust debug syntax. The round-25 formatter didformat!("{:?}", first.kind).split('(').next()onType { kind: Single(JsonType::String) }, which produced “at /: Type { kind: Single” (mismatched brace, cryptic). Now it switches on theValidationErrorKindvariant and emits human-readable messages: “at /: expected type string”, “at /name: required field missing: id”, etc. Falls back tojsonschema’sDisplayimpl for the long-tail kinds. Three new unit tests cover the common scenarios. - [DevX] HTML report removes the duplicate
Negatives by categorysection (#79 round 26 / Srikanth d2). The standalone family rollup is gone; a singleNegatives by categorytable now carries a Family column prepended, with per-row PASS/FAIL badges and clickable mismatch counts (#miss-cat-<cat>anchors). One table, both family + category resolution, no redundancy.
[0.3.169] - 2026-06-05
Fixed
- [Contracts] TUI Conformance tab: selecting a row no longer jumps to a different request when new traffic arrives on the next refresh tick (#79 round 25 / Srikanth follow-up on r24). The 5-second tick was replacing
self.violations, butTableState.selectedwas a numeric index, so the cursor stayed at the same row position while the row’s underlying violation changed. The refresh now re-anchors selection by a(timestamp, method, path, status)identity key (and(timestamp, method, path)for the Unknown view), so the cursor stays on the SAME violation the user clicked. Falls back to position-based behaviour when the prior selection has been evicted by the buffer cap. - [DevX] Capture HTML viewer no longer hangs the browser at 9000+ probes (#79 round 25 / Srikanth d follow-up). Three changes: (1)
content-visibility: autoon every card (browser only paints what’s in the viewport), (2) 200 ms debounce on the search-box input so typing doesn’t trigger a full DOM walk per keystroke, (3) hard cap at 1000 rendered cards with aShowing N of Mbanner pointing users to the JSONL for the full set. The 9700-probe capture from Srikanth’s vCenter run was the trigger.
Added
- [Contracts] Content-type swap probe family
request-body:content-type-mismatch:<variant>(#79 round 25 / Srikanth k variant a). For operations declaring a JSON request body, the self-test driver now sends four probes per operation that lie about Content-Type while keeping the JSON payload:application/xml,application/yaml,multipart/form-data, andapplication/x-www-form-urlencoded(the URL-encoded variant Srikanth added in his round-23 g follow-up). A spec-compliant server should return 415 (or 400). A 2xx means the server is silently accepting payloads outside its declared content negotiation. Skips operations without a JSON body. Required reordering theContent-Typeheader pass insend_case_with_extra(body’s default now runs FIRST so extra_headers’ override wins last) so the probe can actually transmit the wrong content type. Regression testcontent_type_swap_probes_fire_for_json_bodiescovers the variant set and the body-less-op skip. - [Contracts]
--validate-response-schemasflag onmockforge bench --conformance-self-test(#79 round 25 / Srikanth a2 + a3 / closes round 21.3). When set alongside--conformance-self-test-capture, every probe’s response body is validated against the spec’s response schema for the ACTUAL status returned (not just 200), via a per-statusresponse_schemas: BTreeMap<u16, Value>map populated from the OpenAPI annotator. Mismatches land in a newresponse_schema_errorfield on eachCaseCapture(serialised in the JSONL, surfaced as a red “Response schema mismatch” section in the per-probe HTML viewer card). Validation is opt-in because JSON-Schema validation of large response bodies adds wall-clock time on big specs. - [DevX] HTML report gets two new views (#79 round 25 / Srikanth d remainder). (1) A
Negatives by category familyrollup table that groups related categories under a family name (Request body/Parameters/Security family=security+owasp). Family membership is hard-coded so adding a new probe family won’t surprise users with a relabelled row until the map is updated; the detailed per-category table above keeps the full resolution. (2) ABy categorycolumn on the per-operation results table showing the mismatch breakdown per op (e.g.parameters:2 security:1), so a reader can see which probe families an operation fails without expanding the drill-down.
[0.3.168] - 2026-06-04
Fixed
- [Contracts] Geo-source-IP headers now ride on every self-test probe, not just the positive case (#79 round 24 / Srikanth f). Four negative-probe call sites (schema mutation, uri-too-long, missing-header, OWASP injection) and the security-probe path were passing
op.header_paramsdirectly, so theeffective_op_headersgeo-IP append got dropped. Confirmed via Srikanth’s round-23 capture: positive probes carriedX-Forwarded-For: 203.0.113.0, negatives showedgeo={}. A new regression test (geo_headers_present_on_every_probe_with_capture) asserts that every captured probe carries one of the configured forwarded-IP headers. - [DevX] Clickable count cells in the HTML report no longer dead-end when
--report-missed-capcrops the drill-down (#79 round 24 / Srikanth e). The category and per-operation tables now consult a pre-computed anchor set: counts are linked only when their target#miss-cat-*/#miss-op-*row actually survives the truncation, otherwise the count renders as plain text.
Added
- [DevX] Browser-viewable HTML alongside the JSONL capture (#79 round 24 / Srikanth d).
--conformance-self-test-capturenow emits bothconformance-self-test-requests.jsonl(existing, grep-able / streamable) ANDconformance-self-test-requests.html(new, self-contained: open in a browser, no external CSS/JS, includes a toolbar with full-text filter and PASS / FAIL / error checkboxes, one expandable card per probe showing request headers + body, response headers + body, and any transport error). Closes Srikanth’s “Currently json line is plain text, not easy to parse” follow-up.
Notes
- Response-body shape validation (round 21.3 / a2 / a3) and the per-category column / category-family grouping view in the Per-operation table remain queued. The (k) content-type swap probes are explicitly queued now with URL-encoded added to the swap set (Srikanth’s g follow-up).
[0.3.167] - 2026-06-03
Fixed
- [Contracts][DevX]
--source-ipnow actually works with the k6 path (#79 round 23 / Srikanth correction on round-22 g1). The round-22 warning that said “k6 cannot bind a VU to a source IP from the script side” was wrong: k6 supports--local-ipsnatively (CIDR, ranges, and comma-separated single IPs). The k6 executor now forwards the CLI’s--source-ipstraight tok6 run --local-ips, and the misleading warning is removed. Only the--conformance-self-test --use-k6combo still fires a warning, because self-test returns before k6 launches and--use-k6is a no-op there. - [DevX] Docs site is rebuilding again (
docs.mockforge.devhad been stuck since the 2025-11-10 build because mdbook-toc 0.15.x stopped parsing mdbook’s preprocessor input). The TOC preprocessor was disabled (no page uses the<!-- toc -->marker anyway) and itscargo installstep was removed from the deploy workflow, so the round-21 probe-label reference page resolves athttps://docs.mockforge.dev/reference/conformance-self-test-probes.html.
Added
- [Contracts][DevX]
--conformance-self-test-captureflag (#79 round 23 / Srikanth c-iii deferred from round 22.5). When set alongside--conformance-self-test, every probe records method, URL, request headers/body and response status/headers/body toconformance-self-test-requests.jsonl(one JSON object per line) next to the JSON/HTML report. Bodies cap at 16 KiB per direction withrequest_body_truncated/response_body_truncatedflags. Confirms geo-source-ip headers actually shipped and explains why a negative probe came back 200 instead of 4xx, without re-running underRUST_LOG=trace. - [DevX] HTML conformance report: count-cells in the “Negatives by category” and “Per-operation results” tables are now clickable links into the drill-down table below (#79 round 23 / Srikanth d).
#miss-cat-<category>jumps to the first row of that category;#miss-op-<method>_<path-slug>jumps to the first row for that operation. - [DevX] HTML conformance report wording: “missed/caught” renamed to “Mismatched (non-4xx) / Matched (4xx)” across the cards, category table, per-operation table, and drill-down heading; the category status badge is now a plain
PASS/FAIL(replacing “all caught” / “rejection gaps”) since the count column already conveys magnitude (#79 round 23 / Srikanth d wording).
Notes
- Response-body shape validation (round 21.3 / Srikanth a2 / a3) and the per-category column / category-family grouping view in the Per-operation table remain queued for a future round; this release prioritises the regressions and unblockers in Srikanth’s round-22 review.
[0.3.166] - 2026-06-02
Fixed
- [Contracts] TUI Conformance tab: pressing Enter while toggled to the Unknown view (
u) now opens the Unknown Path Detail modal instead of the (wrong) Violation Detail (#79 round 22.1 / Srikanth (f)). The ‘u’ toggle switched the underlying data buffer but the Enter handler routed unconditionally to the violation modal; now both the handler’s “has rows” check and the modal title and body are view-aware.selected_unknown()andselected_unknown_detail()are the new symmetric accessors. - [Contracts][DevX]
--source-ipand--geo-source-ipnow emit a hard warning when set alongside--use-k6or the default--conformancek6 path (#79 round 22.2 / Srikanth (g1)). k6 cannot bind a VU to a source IP from the script side, so--source-ipsilently took only the first IP and never rotated. The warning replaces the silent partial behaviour with an explicit explanation and points users at--conformance-self-test(the native driver) which honours the source-IP pool. A separate warning fires for--geo-source-ipuntil round 22.3 lands (now in the same release).
Added
- [Contracts][DevX]
--geo-source-ipheaders are now rotated through the k6 template too (#79 round 22.3 / Srikanth (g2)). Pre-round-22.3, the geo-IP rotation only applied to the self-test driver; the k6 bench path silently ignored it. The rendered k6 script now declaresGEO_SOURCE_IPS+GEO_SOURCE_HEADERSconstants and merges a rotating__geoHeadersobject into every request’srequestHeadersvia spread.K6Configgainsgeo_source_ipsandgeo_source_headersfields;K6ScriptTemplateDatagainshas_geo_source+ JSON-serialised siblings for the template’s triple-brace embedding. Default header set (when only--geo-source-ipis passed) is the standard three:X-Forwarded-For,True-Client-IP,CF-Connecting-IP. - [Contracts][DevX]
--source-ipand--geo-source-ipnow acceptstart-endIPv4 range syntax alongside CIDR and comma-separated lists (#79 round 22.4 / Srikanth (h)). Pass--source-ip 10.0.0.5-10.0.0.27for 23 hosts without finding a clean prefix. Same 256-host cap as CIDR; backwards ranges (end < start) are rejected with a warning. IPv6 ranges are intentionally rejected because the:separator would collide with the address literal; use CIDR for IPv6. - [DevX] HTML conformance report header now links to the probe-label reference page in the docs (#79 round 22.6 first slice). The “rejection gaps” wording replaces the vaguer “gaps” label on the category status badge, so the badge names what it actually measures.
Notes
- The fuller HTML usability sweep (clickable links from missed counts to drill-down anchors, per-category column in the Per-operation table, category-family grouping view) is queued for round 23 / v0.3.167 alongside response-body shape validation (carried over from round 21.3).
[0.3.165] - 2026-06-01
Added
- [DevX]
--report-missed-cap Nflag onmockforge bench --conformance-self-test(#79 round 21.1) — controls the HTML conformance report’s “Missed negatives” drill-down. Defaults to 200 rows (preserves prior behaviour); pass0for no cap to dump every missed probe into the HTML. The JSON report always carries the full set regardless; this flag only sizes the HTML drill-down so a 50 000-violation run does not produce a multi-megabyte browser-choking file by default. - [DevX] HTML missed-negative table now carries an Expected column (#79 round 21.1) — each row tells you the expected status range (
2xx-3xx (accept)for positives,4xx (reject)for negatives) alongside the Actual status, so the report self-explains without the reader having to remember which probes are 4xx-expecting. Addresses Srikanth’s (a1) ask from round 18. - [DevX] New book page Conformance Self-Test Probes (
reference/conformance-self-test-probes.md, #79 round 21.2) — canonical reference for every probe label the self-test driver can emit. Covers request-body, parameters, security, OWASP, plus howpassedis decided, bucket totals interpretation, and the HTML drill-down cap. Surfaced under Reference inSUMMARY.md.
Notes
- Response-body shape validation alongside the response-code check (Srikanth’s a2 / a3 follow-up) is deferred to a separate release. It needs a non-trivial design pass (multi-status response schema map, body-size cap to avoid OOM on large responses) and was held back to keep this release focused on the HTML UX and docs.
[0.3.164] - 2026-06-01
Fixed
- [Reality][Contracts] Shadow mode now gates its
200response on the configured--base-path(#79 round 20). Pre-round-20,mockforge serve --shadow --base-path /apireturned200 {"shadow":true,"matched":false}for any unmatched path, including ones outside the configured prefix (Srikanth’s report: client used/api123/...against a server configured for/apiand saw200instead of404). Nowdynamic_mock_fallbackcallspath_in_base()which checks at the segment boundary (/api123is NOT under/api), and only returns the shadow200when the path is actually under the configured surface. Outside the surface still 404s in shadow mode, matching pre-shadow semantics. The CLI’s--base-pathflag propagates to the management layer via a newMOCKFORGE_API_BASE_PATHenv var;ManagementState::new()reads it andManagementState::with_base_path()is exposed for callers that want to set it programmatically. Empty or/paths normalise to “no prefix gate” (the pre-round-20 behaviour, useful when neither client nor server has a base path).
[0.3.163] - 2026-06-01
Fixed
- [Contracts] Live-server request-body validator now resolves nested
$refpointers against the spec’s components map (#79 round 19) — Srikanth’s vCenter 0.3.162 run still produced 120"Failed to create schema validator: Pointer '/components/schemas/Esx.Settings.Inventory.EntitySpec' does not exist"violations against schemas that DO exist in the spec. Round 18.3 fixed the bench-side + thevalidation::validate_request_bodyhelper, but missed a third call site inopenapi_routes.rs::validate_request_with_all(line 1250 + 1267) that usedOpenApiSchema::new(...).validate()with a naked validator. Switched both branches toschema_ref_resolver::build_validator(&schema, &spec)which inlines the components. Dotted-name schemas (Esx.Settings.Inventory.EntitySpec,Vapi.Std.DynamicID, etc.) now resolve correctly.
Added
- [Contracts][DevX]
--source-ipand--geo-source-ipnow accept CIDR ranges (#79 round 19 — Srikanth’s follow-up on round 18.5 (j)). Pass--source-ip 10.0.0.0/28to bind 16 hosts,--geo-source-ip 2001:db8::/126for 4 IPv6 hosts, or mix all three forms in one CLI value:--geo-source-ip 10.0.0.0/30,2001:db8::1,203.0.113.42. CIDR expansion is capped at 256 hosts per range to guard against/8typos blowing up the bench client; the cap is logged when triggered. IPv4 and IPv6 supported.
[0.3.162] - 2026-05-31
Issue #79 rounds 17.1 through 18.5 — TUI clipboard + non-violating counter, schema-driven negatives, security probes, spec-level audits, OWASP/WAF unification, HTML report, base-path bug, schema-$ref resolution, OWASP coverage hints, GEODB multi-source-IP testing. Squash-merged from PRs #729 / #731 / #732 / #736 / #737 / #738 / #740 / #741 / #742 / #744. Version chain compressed from incremental 0.3.153–0.3.161 into a single 0.3.162 release; the intermediate version numbers were never published to crates.io.
Added — TUI / Admin
- [Contracts][DevX] TUI Conformance detail view gains a
ckeystroke to copy the selected violation to the system clipboard as pretty-printed JSON (round 17.1) — Srikanth’s (c-i) ask. Usesarboardwith default features so it works on X11, Wayland, macOS NSPasteboard, and Win32 without per-platform setup. On a clipboard-less TTY the failure surfaces in the flash strip instead of silently no-op’ing. - [Contracts] New
total_oklifetime counter on the conformance violations API (round 17.1) — Srikanth’s (f) follow-up. Each request that passes the validator bumps the counter; the admin API returnstotal_okalongsidetotal_seen, and the TUI title now reads e.g.Conformance Violations (256 buffered, 256 shown, 517498/522830 validated failed). Gives the real pass/fail ratio under sustained traffic instead of just the violation count.
Added — Self-test
- [Contracts][DevX] Schema-driven request-body mutator for
--conformance-self-test(round 17.2) — when both a positive sample AND a resolved request-body schema are available, the self-test now emits per-field negatives (type mismatch, required-field removal, min/max bound breaks, pattern miss, enum out-of-range, additional-property), plus a URI-too-long parameter probe. Labels carry the field path (e.g.request-body:type-mismatch:user.email). Bounded bySCHEMA_MUTATION_CAP = 12per operation (top 20 properties, top 5 required) so a 100-property body on a 22 000-operation spec doesn’t produce a runaway test matrix. - [Contracts][DevX][Security] Security probes (round 17.3) — operations declaring a security requirement now get bad-credential negatives (
security:bad-bearer,security:bad-basic,security:bad-apikey:<name>,security:no-auth) plus auth-stripping so the probe’s credential is the only thing the server sees. Surfaces validators that don’t enforce auth even when the spec says they should. - [Contracts][DevX] Spec-level audit alongside
--conformance-self-test(round 17.4) — pure audit of the OpenAPI document (no network I/O) coveringservers(missing/localhost-only/relative-only),callbacks(unsecured webhook ops),polymorphism(oneOf/anyOfwithoutdiscriminator), and a datatype coverage map. Writesconformance-spec-audit.jsonnext to the self-test JSON. - [Contracts][DevX][Security] OWASP / WAF unification into
--conformance-self-test(round 17.5) — folds one canonical payload per OWASP category (owasp:sqli,owasp:xss,owasp:command-injection,owasp:path-traversal,owasp:ssti,owasp:ldap-injection,owasp:xxe) into the existing self-test driver. Injects into the first query param or first string body field; skips operations with no injectable surface. 7 probes max per operation; server should return 4xx (5xx = crashed on payload). - [Contracts][DevX] Self-contained HTML conformance report (round 17.6) —
--conformance-self-testwritesconformance-report.htmlnext to the JSON. Inline CSS, no external assets. Sections: headline cards, negatives by category, per-operation results, missed-negative drill-down (capped at 200 rows), and an optional spec-audit section when a round-17.4 audit JSON is present.
Added — GEODB multi-source-IP
- [Contracts][DevX] GEODB / multi-source-IP testing in
--conformance-self-test(round 18.5).--source-ip <IP>(repeatable) builds a pool of reqwest clients bound vialocal_address();--geo-source-ip <IP>rotates fake source IPs acrossX-Forwarded-For/True-Client-IP/CF-Connecting-IP(configurable via--geo-source-header). Self-test only for v0; bench / k6 path lands in a follow-up.
Changed
- [Contracts][DevX] OWASP coverage table now appends an “Untested OWASP categories” footer that tells you exactly which
--conformance-categoriesvalue to add for each uncovered OWASP category (round 18.4). Removes the “Not Working” confusion when a user-selected subset of categories doesn’t cover the full OWASP Top 10.
Fixed
- [Contracts][DevX]
--conformance-self-testnow honours--base-path(round 18.1) — pre-fix every positive 404’d on deployments served behind a path prefix. - [Contracts][DevX] Self-test now emits a hard warning when every positive case fails with the same status code, instead of silently treating 404s as “negatives caught” (round 18.1).
- [DevX] Bench header no longer prints
Operations: 0 endpointsbefore the spec is parsed (round 18.2). Shows(analyzing spec…)until the count is known. - [Contracts] Request-body validator now resolves nested
$refpointers against the spec’s components map (round 18.3) — pre-fix specs whose component schemas had dotted names (Vcenter.VM.DiskCloneSpec) or were referenced from nested schemas failed withPointer '/components/schemas/X' does not exist. Newschema_ref_resolver::build_validator(&schema, &spec)inlines the components into the validator’s document context.
[0.3.152] - 2026-05-28
Changed
- [Contracts][DevX] TUI Conformance export (
e) now deduplicates by(method, path, category, reason)and sorts by occurrence count descending (#79 round 16) — Srikanth’s (c-ii) ask: under sustained traffic the export was 256 mostly-identical rows. Each group now carries acountandfirst_seen/last_seenwindow so the time range is preserved while the file collapses from “256 rows of mostly the same thing” to a ranked unique-violation list. The TUI table itself still shows each occurrence (unchanged). The flash message now readsexported N unique violation(s) (M occurrences) to …. - [DevX]
mockforge bench --conformance --conformance-self-testnow produces negatives for two previously-zero-coverage shapes (#79 round 16 (h)) — addresses Srikanth’s “always all passing” report on operations that fell through the gaps:- Operations with a required body but no synthesised sample — the
request-body:emptyandrequest-body:wrong-typenegatives now fire whenever the spec declares a request body content type, regardless of whether the body annotator could produce a positive sample. - Operations with a path-param and no required body / query / header — emits a new
parameters:bad-path-paramprobe that substitutes the first path parameter with"self-test-invalid-id". Operations whose spec types the param as integer / UUID / regex-patterned will catch this (4xx); operations that allow free-form strings will let it through. Either outcome is informative — a category that’s all-missedis the spec telling you path-param types are loose.
- Operations with a required body but no synthesised sample — the
[0.3.151] - 2026-05-27
Added
- [Reality]
mockforge serve --shadowCLI flag (#79 round 15) — Srikanth’s (g) ask: a first-class flag for shadow mode instead of onlyMOCKFORGE_SHADOW_MODE=true, so it can’t be silently forgotten. Sets the env var under the hood; the👻 SHADOW MODE ONstartup banner still prints. - [Contracts][DevX] Lifetime “seen total” counters for conformance violations and unknown paths (#79 round 15) — the ring buffers cap at 256 entries, which made a 656k-request run look like “only 256”. Both the admin API (
total_seenfield) and the TUI titles now show the true lifetime count alongside the buffered count (e.g.Conformance Violations (256 buffered, 256 shown, 4821 seen total)). Answers Srikanth’s (f) question.
Changed
- [DevX] Server-side per-violation debug log (#79 round 15) — each recorded conformance violation now emits a
tracing::debug!line under targetmockforge::conformancewith method/path/status/category/reason. Enable precisely withRUST_LOG=mockforge::conformance=debugto get grep-able server-side logs of why each request was rejected, without turning on firehose debug logging. Srikanth’s (b)/(d) ask for “why is this a violation” via logs. - [DevX] TUI Conformance tab readability (#79 round 15) — the
Enterviolation-detail view and the Top Offending Endpoints panel now wrap long lines, so big Microsoft Graph paths and validation reasons are fully visible instead of clipped at the right edge. Srikanth’s (c) ask. (Enterfor full detail,j/kto scroll.)
[0.3.150] - 2026-05-26
Fixed
- [Reality] Server no longer OOM-killed at startup on large specs (#79 round 14) — building the OpenAPI router cloned the entire routes Vec into every per-route handler closure (each closure captured its own
clone_for_validation()), making router construction O(N²) in memory: ~260 GB resident for an 11,422-operation spec (Microsoft Graph), which the OOM killer reaped immediately after theStored N routeslog line. Srikanth hit this on 0.3.148/0.3.149. Fixed by sharing a single validator across all handlers viaArc(per-closure cost drops from a full deep clone to an 8-byte pointer). Applies to all three router builders (build_router_with_context,build_router_with_mockai,build_router_with_ai). Reproduced with a 22,000-operation synthetic spec: pre-fix died withmemory allocation failedright afterStored 22000 routes; post-fix starts and serves cleanly.
Added
- [Reality][Contracts] Server-side shadow mode (
MOCKFORGE_SHADOW_MODE=true) (#79 round 14) — Srikanth’s (a) ask. When enabled, the server returns 200 for requests it would normally reject — unknown paths (instead of 404) and spec violations (instead of 400/422) — while still recording them to the conformance + unknown-paths buffers. A “report-only” / monitor mode: replay proxy traffic through MockForge non-blocking and capture every violation for cross-checking, without the rejections breaking the client flow. Unknown paths get a minimal{"shadow":true,"matched":false}200 stub; spec-violating requests fall through to normal response synthesis. Startup prints a👻 SHADOW MODE ONbanner so the behavior is never a surprise. - [Contracts][DevX] Status column on the TUI unknown-paths view (#79 round 14) — the unknown-paths feed now carries the HTTP status the server actually returned (404 normally, 200 in shadow mode), surfaced as a colored
Statuscolumn in theu-toggled unknown-paths view so you can tell at a glance which requests shadow mode let through.
[0.3.149] - 2026-05-25
Added
- [Contracts][DevX]
mockforge bench --conformance --conformance-self-test(#79 round 13 (4)) — positive + per-category negative driver against a live server. For each annotated operation in--spec: sends one valid request (expect 2xx) plus per-category negatives (empty body, wrong-type body, missing required query param, missing required header) and reports how many the server correctly rejected with 4xx. Writesconformance-self-test.jsonto the output dir; prints a one-line-per-category pass/fail summary; emits a warning when any negative case slipped through. Useful for verifying the round-13 (3) validator-bypass fix took effect against the user’s spec.
[0.3.148] - 2026-05-25
Fixed
- [Contracts][DevX] Conformance buffer now actually fires on the default-flow handlers (#79 round 13) — v0.3.145–0.3.147 wired the buffer infrastructure but the recording call only ran from
build_router_with_context. The MockAI handler (build_router_with_mockai) and AI generator handler (build_router_with_ai) — both of which serve every request when their respective backends are enabled — bypassed validation entirely, so violations never populated for the default-flow routes Srikanth was hitting. Extracted the validate-then-record block intoOpenApiRouteRegistry::run_validation_with_recordingand called it at the entry of each handler closure. Confirmed withcurl -X POST .../users -d '{}'against the demo spec now returning HTTP 400 with the violation showing up inGET /__mockforge/api/conformance/violations.
Added
- [Contracts] New
response-shapeviolation category (#79 round 13) — when a request asks for a status code the spec doesn’t define for that operation (e.g. spec only defines 2xx/4xx butX-Mockforge-Response-Status: 200is sent), record the mismatch under categoryresponse-shapeinstead of silently falling back to the default response. Addresses Srikanth’s (c) question. - [Contracts][DevX] New
unknown-pathsfeed + admin endpoint + TUI view (#79 round 13) — separate bounded ring buffer (mockforge_foundation::unknown_paths) tracks requests whose path didn’t match any route in the loaded spec. Exposed atGET /__mockforge/api/conformance/unknown-paths(DELETE clears). TUI Conformance tab gains aukeystroke to toggle between violations view and unknown-paths view; the existing filter / pause / export / clear / top-endpoints controls work for both. Addresses Srikanth’s (a) question — useful for cross-checking a proxy’s path coverage against the server’s loaded spec.
[0.3.147] - 2026-05-25
Changed
- [DevX] TUI
Conformancetab moved beforeVerification(#79 round 12 follow-up) — Verification consumesTabto cycle its internal fields, so Tab-navigating past it required clicking elsewhere to break the focus. Putting Conformance before Verification keeps it reachable via plain Tab from the earlier screens. Srikanth’s ask.
[0.3.146] - 2026-05-25
Fixed
- [DevX] TUI
Conformancetab now appears in the tab strip + admin server exposes the violations endpoint (#79 round 12 hotfix)- Bug 1 — invisible tab. v0.3.145 added
ScreenId::Conformanceto the enum but forgot to pushConformanceScreen::new()ontoApp::new’s screens vec.render_headeriteratesself.screensso it rendered the other 22 tabs and silently dropped the new one. Fixed; also added adebug_assert_eq!(screens.len(), ScreenId::ALL.len())and aapp_screens_match_screen_id_allunit test so any future enum/vec drift fails before tagging instead of after. - Bug 2 — admin server didn’t expose the endpoint. The TUI client polls the admin URL (
http://localhost:9080) at/__mockforge/api/conformance/violations, but v0.3.145 only wired the route intomockforge-http’s management router (which runs on the mock-traffic HTTP port, not the admin port). TUI got the SPA HTML fallback. Added matchingget_conformance_violations/clear_conformance_violationshandlers inmockforge-uiand routed them at the same path on the admin server — both ports now serve it.
- Bug 1 — invisible tab. v0.3.145 added
[0.3.145] - 2026-05-24
Added
- [Contracts][DevX] New
ConformanceTUI screen +/__mockforge/api/conformance/violationsendpoint (#79 round 12)- Server-side counterpart to the bench-side conformance suite. Every incoming request the OpenAPI router rejects for a spec violation (400/422) is now captured into a bounded ring buffer in
mockforge-foundation::conformance_violations, served byGET /__mockforge/api/conformance/violations, and rendered in a new TUI tab that shows method, path, status, category, client IP, and the rejection reason. Lets you cross-check what your proxy thinks happened against what MockForge thinks happened — Srikanth’s ask on Issue #79. - Extras beyond the ask to make the cross-check-with-proxy workflow easier:
m/s/ccycle method / status-band / category filters (status cyclesany → 4xx → 422 → 5xx); useful when 200+ violations need scoping.ppauses the 5-second auto-refresh so the row under your cursor doesn’t move while investigating.eexports the current (filtered) view toconformance-violations-<ts>.jsonin CWD — drop it next to your proxy logs andjqacross both sides.Dclears the server-side ring buffer (DELETE /__mockforge/api/conformance/violations); useful between test runs.- New per-endpoint count panel shows the top offending
METHOD pathpairs in the current filtered view — quickly highlights which endpoints violate most, not just which categories.
- Server-side counterpart to the bench-side conformance suite. Every incoming request the OpenAPI router rejects for a spec violation (400/422) is now captured into a bounded ring buffer in
Fixed
- [DevX]
mockforge bench --conformance --operations 'METHOD,…'now actually filters (#79 round 12)- Srikanth ran
--conformance --operations "GET,POST"and saw DELETE/PATCH exercised anyway.execute_conformance_testnever appliedself.operations(orself.exclude_operations) when annotating spec operations — those flags were silently dropped. Now applied the same way the regular bench path does, including the existing wildcard / method-only syntax. Also:SpecParser::filter_operationsnow accepts the method-only form"GET"(same asexclude_operationsalready supported) instead of requiring"GET /path".
- Srikanth ran
- [Reality] Multi-target bench summary now includes connection + iteration counts (#79 round 12)
- Srikanth’s
--targets-fileruns were missing theConnections opened/Iterationslines that single-target runs surface.AggregatedMetricsgainedtotal_connections_opened/total_iterations_completedplus per-targetConnections: N Iterations: Mlines under each target, and the new aggregate shows aConnection reuse NOT detectedwarning when the cross-target sum > 5 × VUs.
- Srikanth’s
- [Cloud]
pillar_trackingno longer drivessqlx::pool::acquire“slow acquire” spam under load (#79 round 12)- Round 11 silenced per-event WARNs but the analytics-DB pool still saturated (each event spawned a tokio task that waited 30s on the 10-connection pool). Added an in-flight task cap (default 20, 2× the pool size) at the recorder entry point so over-pressure events are dropped immediately and counted toward the existing aggregated WARN. No more
sqlx::pool::acquire: acquired connection, but time to acquire exceeded slow thresholdstorms.
- Round 11 silenced per-event WARNs but the analytics-DB pool still saturated (each event spawned a tokio task that waited 30s on the 10-connection pool). Added an in-flight task cap (default 20, 2× the pool size) at the recorder entry point so over-pressure events are dropped immediately and counted toward the existing aggregated WARN. No more
[0.3.144] - 2026-05-24
Fixed
- [Reality] OpenAPI router accepts request bodies up to 50 MiB by default (was 2 MiB axum default) — closes the “200 OK before all chunk requests arrived” PCAP behaviour Srikanth reported on Issue #79
- Root cause: axum 0.8’s
BytesandOption<Json<Value>>extractors enforce a 2 MiBDefaultBodyLimit. Above that, the body gets truncated and the handler runs without consuming the rest of the request — hyper sends the response and TLS Close Notify while the client is still uploading the body. Srikanth saw this on a 10 MB chunked PATCH: status 200, response sent at TCP seq 1070094 while the proxy was still pushing chunks at 1070094+1434, 1071528+1434, etc. Reproduced locally with a 3 MiB JSON body to the demo spec — body got truncated at ~2 MiB, JSON parse failed mid-stream, response went out, curl logged* HTTP error before end of send, stop sending. - Fix: every
OpenApiRouteRegistry::build_router_*variant now mountsaxum::extract::DefaultBodyLimit::max(...)with a 50 MiB default, configurable viaMOCKFORGE_HTTP_BODY_LIMIT_MB. 50 MiB covers realistic mock-traffic without giving an untrusted client an unlimited memory-fill vector. Set the env var to a smaller or larger number if your specific workload demands it.
- Root cause: axum 0.8’s
- [Cloud]
pillar_trackingno longer floods logs with one WARN per dropped event under load (#79 round 11)- Under sustained
mockforge bench --rps 100load against an admin-enabledmockforge serve, the analytics DB pool saturated and every failed event emittedWARN ... Failed to record pillar usage event: pool timed out. Pillar tracking is best-effort metrics — losing events under load is acceptable, but the WARN-spam was not. Per-event failures are now DEBUG; a single aggregatedpillar_tracking: dropped X events in the last 60s due to analytics-DB pressureWARN fires at most every 60 seconds.
- Under sustained
[0.3.143] - 2026-05-23
Fixed
- [DevX] Republish of 0.3.142 — fixes broken
[email protected]install (#79 round 10 hotfix)cargo install [email protected]failed witherror[E0432]: unresolved import \crate::database::Database`inmockforge-http/src/handlers/threat_modeling.rs:29andlib.rs:2387. Thedatabasemodule was moved tomockforge-intelligencein #611, but twouse crate::database::Database;statements weren't gated behind#[cfg(feature = “database”)], so default-feature builds (which is whatcargo installuses) failed to compile. Fix landed onmain` in #616/#618 but was not in the 0.3.142 tag.- 0.3.142 of the broken crates (cli, http, import, pipelines, proxy, workspace, core, bench, chaos, collab, recorder, registry-server, k8s-operator, vbr, sdk, test, reporting, ui) yanked from crates.io. 0.3.143 republishes the same #79 round-10 changes from a known-good main commit.
[0.3.142] - 2026-05-21
Changed
- [DevX] Pre-flight
--vusrecommendation caps at 1000 for huge specs (#79 round 10)- Srikanth’s 11,422-operation spec at
--rps 100(9.4ms baseline) produced a recommendation of ~10,740 VUs — mathematically correct but practically absurd. The right answer for that workload isn’t “spin up 10K VUs”, it’s “shrink the workload.” 0.3.142 caps the printed recommendation at 1000 and, above the cap, steers users toward--operations/--exclude-operationsfilters or dropping--rpsfor closed-model loading.
- Srikanth’s 11,422-operation spec at
Added
- [DevX] Iteration coverage in bench summary (#79 round 10)
- New
Iterations: X complete × N ops = Y ops fully exercisedline appears whenever k6’siterations.values.countis non-zero. When the run ends mid-iteration (large spec, undersized VUs), a follow-up line surfaces the extra requests from the partial pass so users know the last iteration didn’t cover every operation in the spec.
- New
Fixed
- [DevX]
scripts/check-changelog.shnow uses the PR-wide diff in CI instead ofgit diff-tree -r HEAD(which returns the empty combined diff on the syntheticrefs/pull/<N>/mergecommitactions/checkoutuses). Previously every PR whose CHANGELOG.md edit landed in a commit that the merge didn’t have to reconcile would fail the “Changelog Validation” gate even though the workflow’s own outer condition correctly detected the edit (caught on #595’s release PR). Now switches modes on$GITHUB_BASE_REF: PR-widegit diff --name-only origin/$GITHUB_BASE_REF...HEADin CI, single-commitdiff-tree -r HEADlocally for the cargo-release flow.
[0.3.141] - 2026-05-20
Fixed
- [DevX]
mockforge-foundation::pillarsdoctests now reference the actual crate path (release-gate fix)- All 9 doctests imported
use mockforge_core::pillars::..., but thePillarenum lives inmockforge-foundationitself. The mismatch causedcargo test --doc -p mockforge-foundationto fail withcannot find module or crate `mockforge_core`, which gates the Release workflow’scargo test --workspace --releasestep — so the v0.3.140 tag’sCreate Releasejob failed and Publish-to-crates.io was skipped. Replacedmockforge_core::pillars→mockforge_foundation::pillarsacross all 9 doctests. 11 doctests now pass. - 0.3.140 was never published to crates.io; this release ships the round-9 bench fix originally intended for that version, plus this doctest fix on top.
- All 9 doctests imported
[0.3.140] - 2026-05-20
Fixed
- [DevX] Pre-flight
--vusprobe now factors in operations-per-iteration (#79 round 9)- The round-8 probe assumed 1 iteration = 1 request, but k6’s
constant-arrival-ratecounts iterations and every iteration calls every operation in the spec. Srikanth’s 12-op spec at--rps 100with 15ms latency reported “–vus 5 is sufficient” and then k6 still emitted “Insufficient VUs, reached 5 active VUs” mid-run because real iteration time was 15ms × 12 ops, not 15ms. - Fix:
ProbeResult::required_vus(rps, num_operations)now multiplies by spec operation count. Same call site change incommand.rs. New testsrequired_vus_scales_with_operation_countandrequired_vus_treats_zero_operations_as_one. The pre-flight progress / warning lines now print the operation count so users see what the math used.
- The round-8 probe assumed 1 iteration = 1 request, but k6’s
[0.3.139] - 2026-05-19
Fixed
- [Cloud] Drop unnecessary
std::time::qualifier inincident_dispatcherPagerDuty timeoutDurationis already imported at the top ofcrates/mockforge-registry-server/src/workers/incident_dispatcher.rs, so the fully-qualifiedstd::time::Duration::from_millis(200)at line 698 was tripping the workspace-wide-Dunused_qualificationsratchet. The Incremental Warning Gate started failing on this when PR #573 landed; this clears it without touching behaviour.
Added
- [DevX] Adaptive pre-flight latency probe for
--vussizing (#79 round 8)mockforge bench --rps N --vus Mnow does a 3-request HEAD/GET probe of the target before launch to measure baseline latency, then recommends--vusbased onceil(rps × measured_latency_secs) + 1instead of the static “100ms / 10 req-per-VU” heuristic. Fixes a false-positive on Srikanth’s fast targets (~2ms response time) where the old heuristic warned “bump to –vus 100” when –vus 5 was actually plenty.- If the probe can’t reach the target (auth-gated, strict WAF, etc.), falls back to the previous 100ms heuristic so the warning still fires when warranted. When the probe succeeds AND
--vusis sufficient, prints a confirmation line so users know the size check passed.
- [Reality] “Connection reuse NOT detected” diagnostic in bench summary (#79 round 8)
- When
tcp_connect_samples > 5 × vus_maxin pooled-reuse mode (no--cps), the bench reporter now prints a yellow warning explaining the target is closing sockets between requests. Addresses Srikanth’s 0.3.137 question: “I expected 5 connections with –vus 5 but see 7425” — the counter is correct, the proxy isn’t pooling. The new line makes that interpretation explicit so users don’t have to guess.
- When
[0.3.138] - 2026-05-19
Added
- [Cloud][Reality] Cloud-mode Time Travel — controllable virtual clock on hosted-mock deployments (#466, #527)
- 7 clock-control endpoints proxied (
status,enable,disable,advance,set,scale,reset) over Fly 6PN. Wire format mirrors cloudResilience and cloudWorldState:{ runtime_state: "live" | "unreachable", data }. The runtime-unreachable branch disables the control fieldset so users don’t fire mutations that’ll bounce back. - New
CloudTimeTravelViewreuses the local TimeTravelPage shape. Cron jobs + mutation rules stay local-only — they manage scenario state, not a hosted mock’s single-process clock. - Note: these entries were originally written into the 0.3.137 changelog block — the PRs landed after the v0.3.137 tag (efa43d20) was published to crates.io, so they actually ship in 0.3.138.
- 7 clock-control endpoints proxied (
- [Cloud][AI] Cloud-mode Test Generator — async LLM jobs over runtime_captures (#469, #529)
- New
cloud_test_generation_jobstable + 4 CRUD endpoints + SSE stream (GET .../jobs/{id}/stream) for live progress. - Background tokio worker drains the queue with
FOR UPDATE SKIP LOCKEDclaims, processes up toTEST_GENERATION_WORKER_CONCURRENCY(default 4) jobs in parallel per tick, dispatched via the sameai::client+ai::quotapipelineai_studiouses. BYOK skips the platform token quota; paid plans without BYOK succeed via the platform key. - Worker is forgiving: best-effort JSON parse strips
```jsonfences, recovers from prose wrappers, falls back to a{ raw_content }wrapper. Cancellation race is safe — terminal writes are gated onWHERE status = 'running'. - New
CloudTestGeneratorViewwith job timeline, create form, expandable detail rows, and SSE-driven sub-second updates on expanded non-terminal rows.
- New
- [Cloud][Reality] Real-time runtime-logs SSE via Fly NATS subscription (#556, #559)
- The registry now subscribes to Fly’s NATS log stream and re-broadcasts deployment logs over SSE so the cloud UI can render live logs without polling. Replaces the previous pull-based log endpoint.
- [Cloud][Reality] OTLP gRPC trace receiver alongside HTTP/JSON (#548, #566)
- Registry exposes a standard
:4317OTLP gRPC endpoint in addition to the existing HTTP/JSON path. Hosted mocks can now use OpenTelemetry SDKs configured for either transport.
- Registry exposes a standard
- [Cloud][Reality] Per-capture cloud forwarder with backpressure + retry (#553, #564)
mockforge-recordergained a registry-bound forwarder that streams individual captures to the cloud control plane with exponential-backoff retries and a bounded in-memory buffer for backpressure. Replaces the old end-of-recording bulk upload.
- [Cloud][Reality] Trust-root boot — plugin host fetches + refreshes active trust roots from registry (#549, #565)
- The plugin host fetches the active set of plugin trust roots from the registry on boot and refreshes them on a schedule, so signed manifests written against rotated trust roots verify without a host redeploy.
- [Cloud][Reality] HSM-backed platform signing-root rotation via AWS KMS (#550, #567)
- Platform signing roots can now be rotated end-to-end via AWS KMS without exposing private key material to host code. Replaces the file-backed signing root used in earlier versions.
- [Cloud][Reality] Email notification channel via EmailService (#551, #557)
- Registry’s notification dispatcher gained an Email transport alongside Slack, so cloud incidents/alerts can fan out to email recipients configured per-organization.
- [Cloud][Reality] PagerDuty notification channel via Events API v2 (#552, #558)
- Same dispatcher gained a PagerDuty Events API v2 transport. Incidents created in cloud can now page on-call rotations directly.
- [CI][Reality] Nightly hosted-mocks smoke workflow (#554, #563)
- New
hosted-mocks-smoke.ymlruns nightly against the production Fly deployments to catch regressions that pass per-PR CI but break in cloud. Outputs feed the runtime daemon’s smoke incidents tab.
- New
Changed
- [DevX]
pr_generationmoved out ofmockforge-coreintomockforge-intelligence; the intelligence → core dependency cycle was broken (#562 phase 1, #571)mockforge-intelligencedropped itsmockforge-coredep, freeingmockforge-coreto take amockforge-intelligencedep without Cargo rejecting it. The two real uses (mockforge_core::Resultandmockforge_core::scenarios::ScenarioDefinition) moved tomockforge_foundation::Resultandmockforge-http::handlers::behavioral_cloning.- Backwards compat:
mockforge_core::pr_generationis preserved aspub use mockforge_intelligence::pr_generation;— existing call sites compile unchanged. External callers (mockforge-recorder,mockforge-pipelines,mockforge-http,mockforge-collab) updated to import from the new home.
- [DevX] ADR auditing
mockforge-httpfor intelligence/proxy extraction (#555, #561)- Documentation-only ADR (
docs/adr/0001-mockforge-http-extraction.md) recording the dependency analysis and migration plan that #562 phase 1 acts on.
- Documentation-only ADR (
Fixed
- [DevX] CI rust-cache no longer poisons builds with dangling
target/*.dpaths (#446, #570)- Root cause: every workflow set
CARGO_HOME=/tmp/cargo-mockforge-${{ github.run_id }}(unique per run) so the next run’sSwatinem/rust-cache@v2restored atarget/whose.dfiles referenced the previous run’s CARGO_HOME — long since deleted. Surfaced aserror: could not compile <crate> ... (never executed) No such file or directory (os error 2)on chronically red jobs. - Fix: switched every
CARGO_HOMEtorunner.name(stable per machine) and addedwith: env-vars: "CARGO_HOME"to all 13Swatinem/rust-cache@v2invocations so the cache key partitions by runner.
- Root cause: every workflow set
- [UI][Cloud] CloudTestGeneratorView import path corrected (
stores/useWorkspaceStore) (#547)- Minor follow-up to #529; runtime error on cloud mode without it.
[0.3.137] - 2026-05-18
Added
- [Cloud][Reality] Cloud-mode World State — per-deployment graph + snapshot + layers + slice-query surface (#464, #528)
- Registry proxies 5 HTTP endpoints (
snapshot,snapshot/{id},graph?layers=…,layers,query) over Fly 6PN to{fly-app}.internal:3000/api/world-state/*. Wire format mirrors cloudResilience and cloudTimeTravel:{ runtime_state: "live" | "unreachable", data }.unreachablecarriesdata: nullso the UI renders an honest empty state. - New
CloudWorldStateViewreuses the existingStateLayerPanel,WorldStateGraph, andStateNodeInspectorcomponents so the visualization is identical to local mode. Deployment selector auto-picks the first active hosted-mock; multi-deployment orgs get a dropdown. 5-second polling matches the local TanStack QueryrefetchInterval. 'world-state'added tocloudNavItemIds. WebSocket/streamupstream is intentionally not proxied — polling parity is functionally equivalent for the cadence the local UI uses, and ws-tunneling through 6PN is a follow-up.
- Registry proxies 5 HTTP endpoints (
- [Cloud][Reality] Cloud-mode Time Travel — controllable virtual clock on hosted-mock deployments (#466, #527)
- 7 clock-control endpoints proxied (
status,enable,disable,advance,set,scale,reset). Targets the hosted mock’s main HTTP port (3000) — not the admin port — because the time-travel router mounts there for reachability when admin isn’t publicly exposed. - New
CloudTimeTravelViewwith a runtime-unreachable banner that disables the control fieldset so users don’t fire mutations that’ll bounce back. Cron jobs + mutation rules stay local-only — they manage scenario state, not a hosted mock’s single-process clock. 'time-travel'added tocloudNavItemIds.
- 7 clock-control endpoints proxied (
- [Cloud][AI] Cloud-mode Test Generator — async LLM jobs over runtime_captures (#469, #529)
- New
cloud_test_generation_jobstable + 4 CRUD endpoints + SSE stream (GET .../jobs/{id}/stream) for live progress. - Background tokio worker drains the queue with
FOR UPDATE SKIP LOCKEDclaims, processes up toTEST_GENERATION_WORKER_CONCURRENCY(default 4) jobs in parallel per tick, and dispatches via the sameai::client+ai::quotapipelineai_studiouses — so quota and billing semantics stay consistent. BYOK skips the platform token quota; paid plans without BYOK succeed via the platform key (MOCKFORGE_PLATFORM_LLM_API_KEY+ provider/model/endpoint env vars). - Worker is forgiving: best-effort JSON parse strips
```jsonfences, recovers from prose wrappers, falls back to a{ raw_content }wrapper. Cancellation race is safe — terminal writes are gated onWHERE status = 'running'so a user-cancel mid-flight wins. - New
CloudTestGeneratorViewwith job timeline + create form + expandable detail rows + SSE-driven sub-second updates on expanded non-terminal rows.'test-generator'added tocloudNavItemIds.
- New
- [DevX] Pre-flight warning when
--vusis too low for--rps(#79 round 6 follow-up, #543)mockforge bench --rps N --vus Mnow warns before launch whenM × 10 < N(rule of thumb: 1 VU at ~100ms latency sustains ~10 req/s). The warning suggests a higher--vusvalue (ceil(rps / 10)), so users hit by k6’s “Insufficient VUs, reached M active VUs and cannot initialize more” message know what to change.
Changed
-
[CI][Reality]
release.ymlgates Fly deploy + crates.io publish + Helm chart push on thereleasejob’scargo test --workspace --release(#447, #526)- All three downstream jobs (
deploy-registry,publish,helm) nowneeds: release. A tag with red tests no longer ships to Fly / crates.io / the Helm repo. Cost is a few minutes of latency on a green deploy; benefit is no more “release auto-deploys broken code” surprises.
- All three downstream jobs (
-
[Architecture]
ai_studio(the final AI cluster module) moved whole tomockforge-intelligence(Issue #562 phase 8 — campaign complete)- 17 files / ~7,200 LOC:
api_critique,artifact_freezer,behavioral_simulator,budget_manager,chat_orchestrator,config,contract_diff_handler,conversation_store,debug_analyzer,debug_context,debug_context_integrator,nl_mock_generator,org_controls,org_controls_db,persona_generator,system_generator,mod. - All inter-AI deps (
ai_contract_diff,contract_validation,failure_analysis,intelligent_behavior,reality,voice) became sibling modules inmockforge-intelligenceafter phases 2–7, socrate::*paths in the moved code resolved without rewriting. Only externals needing swap:crate::{OpenApiSpec, Result}→mockforge_foundation::Result+mockforge_openapi::OpenApiSpec, andcrate::pillar_tracking→mockforge_foundation::pillar_tracking. - Backwards compat:
pub use mockforge_intelligence::ai_studio;inmockforge-core/src/lib.rs. Allmockforge_core::ai_studio::*callers (mockforge-http handlers, mockforge-ui handlers) keep compiling unchanged. mockforge-intelligencegaineddirs,once_cell,sqlx(optional under newdatabasefeature).mockforge-core’sdatabasefeature now activatesmockforge-intelligence/databasetoo, so thedatabase-gatedorg_controls_dbsqlx surface continues to work end-to-end.- Issue #562 complete. All 4 AI cluster modules from the original deferral note are now in
mockforge-intelligence:ai_contract_diff,intelligent_behavior,threat_modeling,ai_studio. The 7-phase campaign also re-homedpillars,pillar_tracking,chaos_utilitiestomockforge-foundationand migrated supporting modulescontract_validation,failure_analysis,reality, plus 6 of 7voicefiles (the 7th,voice_workspace, stays in core because its multi_tenant/scenarios/workspace deps would require their own multi-day extractions).
- 17 files / ~7,200 LOC:
-
[Architecture]
voicesplit: 6 leaf files moved tomockforge-intelligence,workspace_builderstays in core (Issue #562 phase 7)- The wall:
voice(3,769 LOC across 7 files) was the last AI cluster blocker. Scoping showed all heavy deps (multi_tenant,scenarios,workspace,contract_drift,reality_continuum) are concentrated in a single file (workspace_builder.rs, 536 LOC). The other 6 files (3,233 LOC) only depend onmockforge-openapi+ siblingintelligent_behavior+mockforge-foundation. - Split, not whole-move:
command_parser,conversation,hook_transpiler,spec_generator,workspace_scenario_generator,mod→mockforge_intelligence::voice.workspace_builderrenamed tomockforge_core::voice_workspace(stays in core; needs the heavy deps and the imports go back throughcrate::voice::*shim). - Backwards compat: new
mockforge_core::voiceshim consolidates both halves withpub use mockforge_intelligence::voice::*; pub use crate::voice_workspace::{BuiltWorkspace, WorkspaceBuilder};plus sub-module re-exports (command_parser,spec_generator, etc.) so existing path-style imports keep working.mockforge-uihandler updated one import (workspace_builder::WorkspaceBuilder→WorkspaceBuildersince it now comes through the top-level shim). - Brings
ai_studioto 0 dirty sub-files. Phase 8 will move ai_studio whole.
- The wall:
-
[Architecture]
realitymoved tomockforge-intelligence,chaos_utilitiesre-homed tomockforge-foundation(Issue #562 phase 6)reality.rs(541 LOC) holdsChaosConfigas a struct field — moving it to intelligence requiredChaosConfigto live in a crate intelligence can depend on (foundation).chaos_utilities.rs(500 LOC) only depends onfailure_injection+latency, both already in foundation, so the move is mechanical.- The
RealityEngine::apply_to_configinherent method (which pokes at concreteServerConfigsub-structs) moved tomockforge_core::reality_apply::apply_reality_to_server_configas a free function. Keeping it onRealityEnginewould have forced intelligence → core dep again. CLI updated to use the new helper (one call site). - Backwards compat:
mockforge_core::{chaos_utilities, reality}preserved aspub use. ExistingRealityEngine,RealityConfig,RealityLevel,ChaosConfig, etc. continue resolving viamockforge_core::*for all 5+ external callers (cli, http, ui, registry, scenarios, etc.). - Brings
ai_studioto 1 dirty sub-file:nl_mock_generator.rs(depends onvoice). After phase 7 extracts voice,ai_studiomoves whole in phase 8.
-
[Architecture]
contract_validation+failure_analysismoved tomockforge-intelligence(Issue #562 phase 5)- Both modules are tiny leaves with near-zero core coupling:
contract_validation.rs(590 LOC, single file, onlyserdenon-internal deps) andfailure_analysis/(4 files / ~512 LOC, only depends on siblingintelligent_behavior). Inlinecrate::openapi::OpenApiSpecandcrate::pillar_tracking::record_contracts_usagereferences incontract_validationswapped to their home crates (mockforge_openapi/mockforge_foundation::pillar_tracking). - Backwards compat:
mockforge_core::{contract_validation, failure_analysis}preserved aspub use mockforge_intelligence::*;. Zero churn for 4 external callers (mockforge-ui handlers, mockforge-cli, in-core ai_studio + workspace). - Brings
ai_studioone step closer to a clean whole-module move: 5 dirty sub-files → 3 dirty (debug_context + debug_context_integrator still needreality; nl_mock_generator still needsvoice).
- Both modules are tiny leaves with near-zero core coupling:
-
[Architecture]
ai_contract_diffmoved tomockforge-intelligence,pillars+pillar_trackingre-homed tomockforge-foundation(Issue #562 phase 4)- Phase 3 deferred
ai_contract_diffbecause itsDiffAnalyzer::analyze_with_recommendationscalledcrate::pillar_tracking::record_ai_usage— an analytics global living inmockforge-corethat would have re-introduced an intelligence → core dep. Phase 4 unblocks it by promotingpillars.rs(568 LOC) andpillar_tracking.rs(207 LOC) tomockforge-foundation. Both files were already self-contained (only depend onserde/chrono/once_cell/async_trait/tracing), so the move is mechanical. - With pillar_tracking in foundation,
ai_contract_diff(7 files / ~2,380 LOC) moves cleanly tomockforge-intelligence. The singlepillar_tracking::record_ai_usage("ai_generation", {"type":"contract_diff",...})call site is preserved — analytics dashboards keep receiving contract-diff usage events unchanged. - Backwards compat:
mockforge_core::pillars,mockforge_core::pillar_tracking, andmockforge_core::ai_contract_diffare all preserved aspub usere-exports. Every existing call site — 10+ files usingpillar_tracking(ui, registry-server, scenarios, http, cli, ai_studio, voice, contract_validation, workspace) and the 5+ files usingai_contract_diff(http, cli, in-core consumers) — keeps compiling unchanged. mockforge-foundationgainedonce_cell+tracingworkspace deps (both already used by every dependent crate; this just makes the dep direct).ai_studioremains in core — needsreality,voice,failure_analysis, andcontract_validationextracted first (12 of 17 sub-files are otherwise clean post-phase-4, but a partial split would fracture module identity). Documented inmockforge-intelligence/src/lib.rs.
- Phase 3 deferred
-
[Architecture]
threat_modeling(security analyzers) moved out ofmockforge-core::contract_driftintomockforge-intelligence(Issue #562 phase 3)- 8 files / ~1,300 LOC:
dos_analyzer,error_analyzer,pii_detector,remediation_generator,schema_analyzer,threat_analyzer,types,mod. - Foreign deps swapped to their home crates:
crate::Result→mockforge_foundation::Result,crate::openapi::OpenApiSpec→mockforge_openapi::OpenApiSpec.crate::intelligent_behavior::*stays ascrate::intelligent_behavior::*because it’s now a sibling module inmockforge-intelligence(phase 2 moved it there). - Backwards compat:
mockforge_core::contract_drift::threat_modelingis preserved aspub use mockforge_intelligence::threat_modeling;frommockforge-core/src/contract_drift/mod.rs. External callers (mockforge-http,mockforge-cli) continue to import via themockforge_core::contract_drift::threat_modelingpath — zero caller-side churn. ai_contract_diffwas scoped out of this phase because itsDiffAnalyzer::analyze_with_recommendationscallscrate::pillar_tracking::record_ai_usage, which is an analytics global living inmockforge-coreand would either need re-homing or a callback-injection refactor to move cleanly. Documented as a phase-4 prerequisite inmockforge-intelligence/src/lib.rs.
- 8 files / ~1,300 LOC:
-
[Architecture]
intelligent_behavior(AI cluster leaf) moved out ofmockforge-coreintomockforge-intelligence(Issue #562 phase 2)- With the cycle broken in phase 1, the leaf of the AI cluster could move.
intelligent_behavior’s only foreign deps weremockforge_core::Result(nowmockforge_foundation::Result) andmockforge_core::openapi::OpenApiSpec/openapi_routes::OpenApiRouteRegistry/openapi::response::ResponseGenerator— all of which already live inmockforge-openapi(the core paths were just re-exports).mockforge-openapiitself depends only onmockforge-foundation, so the migration stays cycle-safe. - 24 files (~7,800 LOC) moved:
behavior,cache,condition_evaluator,config,context,embedding_client,history,llm_client,memory,mockai,mutation_analyzer,openapi_generator,pagination_intelligence,relationship_inference,rule_generator,rules,session,spec_suggestion,sub_scenario,types,validation_generator,visual_layout. - Backwards compat:
mockforge_core::intelligent_behavioris preserved aspub use mockforge_intelligence::intelligent_behavior;. Every existingcrate::intelligent_behavior::*import insidemockforge-core(voice,reality,graph,ai_contract_diff,ai_studio,contract_drift,failure_analysis) keeps compiling unchanged. External callers (mockforge-chaos,mockforge-cli,mockforge-http,mockforge-ui,mockforge-vbr, workspacetests/) continue to import via themockforge_corere-export — no caller-side churn this phase. mockforge-intelligencegrewmockforge-openapi,indexmap,openapiv3,regex,sha2, andtokiodeps. Per-doctest fix: themockforge_core::intelligent_behaviorusage example inmod.rswas retargeted tomockforge_intelligence::intelligent_behavior.- This unblocks phase-3 follow-ups (
ai_contract_diff,threat_modeling) — each can now also use thepub use mockforge_intelligence::*;shim pattern since theintelligent_behavior::{config, llm_client, types}they all depend on is reachable from the intelligence side.ai_studioremains blocked untilreality/voice/failure_analysis/contract_validationare extracted.
- With the cycle broken in phase 1, the leaf of the AI cluster could move.
-
[DevX]
pr_generationmoved out ofmockforge-coreintomockforge-intelligenceand the intelligence → core cycle was broken (Issue #562 phase 1, #571)- Why this matters: the ADR for #555 (
docs/adr/0001-mockforge-http-extraction.md) and themockforge-intelligence/src/lib.rsdocstring both identified the bidirectionalmockforge-core↔mockforge-intelligencedependency as the blocker behind every other AI submodule extraction. With the cycle broken, future moves (behavioral_economics,ai_contract_diff,ai_studio, etc.) become mechanical — they no longer have to fight the dep graph. - How the cycle was broken:
mockforge-intelligencedropped itsmockforge-coredep entirely. The two real uses (mockforge_core::Resultandmockforge_core::scenarios::ScenarioDefinition) becamemockforge_foundation::Result(zero-cost swap —core::Resultalready re-exportsfoundation::Result) and a one-method move ofSequenceLearner::generate_sequence_scenariointomockforge-http::handlers::behavioral_cloning(its only caller). That freedmockforge-coreto take amockforge-intelligencedep without Cargo rejecting it. - Backwards compat:
mockforge_core::pr_generationis preserved aspub use mockforge_intelligence::pr_generation;. Every existingcrate::pr_generation::*import inside core (e.g.,config::modanddrift_gitops::handler) keeps compiling unchanged. External callers (mockforge-recorder,mockforge-pipelines,mockforge-http,mockforge-collab) updated to import from the new home; themockforge_core::pr_generationpath remains valid for any out-of-tree consumers. mockforge-intelligencegained aschemafeature mirroring core’s (so the migrated types still gate theirJsonSchemaderive the same way) and grew the depspr_generationneeds (base64,reqwest,urlencoding,mockforge-foundation).
- Why this matters: the ADR for #555 (
Fixed
- [DevX] CI rust-cache no longer poisons builds with dangling
target/*.dpaths (Issue #446, #570)- Root cause: every workflow set
CARGO_HOME=/tmp/cargo-mockforge-${{ github.run_id }}(unique per run) so the next run’sSwatinem/rust-cache@v2restored atarget/whose.dfiles referenced the previous run’s CARGO_HOME — long since deleted. Surfaced aserror: could not compile <crate> ... (never executed) No such file or directory (os error 2)on chronically red jobs (Test stable, Incremental Warning Gate, Code Coverage). - Fix: switched every
CARGO_HOMEtorunner.name(stable per machine, still unique across the 7 sibling runners that share the host) so cached dep-info paths remain valid across runs on the same runner. Addedwith: env-vars: "CARGO_HOME"to all 13Swatinem/rust-cache@v2invocations so the cache key partitions by runner — different runners can’t restore each other’s caches and re-introduce the same staleness across the runner pool.
- Root cause: every workflow set
- [Reality] Client-side “Connections opened” counter now appears for
--rps-only runs (#79 round 6 follow-up, #543)- Root cause: the parser was reading
http_req_connecting.values.countfrom k6’ssummary.json, but k6’s Trend metric never emits acountfield — onlyavg/min/med/max/p(90)/p(95). The field was always absent, sotcp_connect_sampleswas always 0 and the connection-count line never printed for non---cpsruns. - Fix: the generated k6 script now declares a dedicated
mockforge_connections_openedCounter and increments it wheneverres.timings.connecting > 0(i.e. a fresh TCP socket was opened). The Rust parser reads this Counter’scountdirectly. Works for both--cpsruns (≈ total requests) and pooled-reuse runs (≈vus_max). - Also: TCP-connect / TLS-handshake timing lines now print whenever the Trend has a non-zero
avg, not whencount > 0(which was unreliable). Newtest_connections_opened_counter_presentregression test guards both the Counter declaration and the per-request increment.
- Root cause: the parser was reading
- [Reality]
--scenario constantnow runs at full VU concurrency from t=0 (#79 round 6 follow-up, #543)- Root cause: Srikanth reported that
--vus 5 -d 600stook until the ~6-minute mark to reach 5 VUs and then ramped DOWN. The k6 template always wrotestartVUs: 0, so even--scenario constant’s single{duration: '600s', target: 5}stage maderamping-vuslinearly interpolate from 0 → 5 across the whole window. - Fix: for
Constant,startVUsis seeded atmax_vusso concurrency is at full from the start. Ramping scenarios (RampUp/Spike/Stress/Soak) still start at 0 and let their stages drive the curve. Guarded bytest_constant_scenario_starts_at_target_vus.
- Root cause: Srikanth reported that
- [Cloud][Reality] Cloud Resilience dashboard could not reach hosted mocks over Fly 6PN (#468 follow-up, #542, #544)
mockforge serve --adminwas binding the admin port to127.0.0.1, so the registry proxy’s{fly-app}.internal:9080reach failed silently (#542 — bind dual-stack). UI’s/api/resilience/*paths also sat behind auth middleware that doesn’t run in proxied cloud mode, so the proxy’s outbound calls 401’d (#544 — explicit exempt prefix). Both fixes were needed for the cloud Resilience tab to render live state instead of perpetualruntime_state: unreachable.
- [UI][Build] Allow
esbuild/vue-demibuild scripts under pnpm 11’s managed-builds policy (#525, #533)pnpm-workspace.yamldeclares the two packages inonlyBuiltDependenciesso production docker builds succeed. Fix split across two commits because the first round (#525) put the allowlist in the wrong file.
- [UI][Build] Pin pnpm to 10.15.0 in the docker stages (#525 follow-up, #536)
- 10.15.0 predates the managed-builds policy that triggered #525 in the first place. Single-pin keeps build behaviour consistent across local + CI + cloud regardless of pnpm’s release cadence.
- [UI][Cloud] Hide
api-explorerfrom cloud sidebar (#459 follow-up, #537)- API Explorer is still reachable from the HostedMocksPage “Open” action, so it doesn’t need a standalone cloud sidebar slot.
- [CI][Docker] Explicit cosign login so signature push to GHCR succeeds (#546)
- GHCR rejected the unsigned cosign push with 401 after PAT-vs-GITHUB_TOKEN cleanup; cosign now logs in explicitly before push.
Cloud-parity meta tracker
This release closes #459 — the cloud-parity meta tracker for the 14 “Local only” nav items. 14 / 14 addressed: 8 shipped end-to-end across prior releases (Graph #460, Virtual Backends #461, Logs #462, Metrics-folded #463, World State #464, Observability #465, Performance-folded #467, Resilience #468), 4 explicitly kept local-only (Proxy Inspector #470, SMTP Mailbox #471, MQTT Broker #472, Kafka Broker #473), 2 land in this release (Time Travel #466, Test Generator #469).
[0.3.136] - 2026-05-15
Added
- [Registry][Security] Usage-limit enforcement +
past_dueread-only mode — closes launch blocker #449 (#515)- 429 spec body for quota exhaustion: new
ApiError::UsageLimitExceeded { limit_type, current, max, period }returns the{"error":"usage_limit_exceeded","limit":"…","current":N,"max":M}shape from the issue. Wired into both the hosted-mock proxy’s inlineenforce_monthly_quotaand the previously-deadorg_rate_limit_middleware, so the same response shape comes out of every path that ever has to reject for quota. Free-tier orgs no longer have an unbounded request budget; Pro/Team orgs trip 429 at their plan ceiling instead of silently consuming Team-tier volume on a Pro plan. past_dueread-only mode after 24h grace: newpast_due_writes_blockedmiddleware on the authenticated route stack. Once an org is inpast_duepast the 24h grace window (introduced in #507), write methods outside an explicit billing/auth/support/legal allowlist return 402 PaymentRequired. Reads + recovery paths (billing portal, support, legal) stay fully reachable so the customer can self-serve out of dunning. Closes the “customer keeps consuming compute through the 7–10 day Stripe retry window” leak.- 402 on past_due deploys (criterion 8): the deploy-time past_due gate now returns 402 PaymentRequired via a new
ApiError::PaymentRequiredvariant, distinct from the old 400 InvalidRequest — billing-state failure is a distinct response class from a malformed request. - Integration coverage: new
crates/mockforge-registry-server/tests/usage_limits_e2e.rsexercises the free→429-with-spec-body path, past_due→402 on the deploy handler, past_due→402 onPOST /api/v1/workspaces(proves the route-wide middleware works, not just the inline check), and confirms reads + billing endpoints stay reachable. Gated#[ignore]for DATABASE_URL availability; run via the registry’s E2E job. - This closes the last 3 of #449’s 8 acceptance criteria; the deploy/workspace/member gates (#479),
requests_per_30dhosted-mock proxy enforcement (#494), and 24h past_due grace window (#507) had already shipped in earlier patches.
- 429 spec body for quota exhaustion: new
Fixed
- [Reality] Drained workspace-wide
unused_qualificationsregressions inmockforge-http::counting_listener(9 sites, from #520) andmockforge-registry-server::handlers::resilience(1 site, from #522) (#515, second commit)- Both PRs ran the warning gate’s scoped check (
mockforge-cli+mockforge-ui) only, so the workspace-wideunused_qualificationsratchet (graduated workspace-wide in #500-#511) caught these on rebase rather than at merge time. Trivial removals of redundanttower::,std::pin::,std::task::, andserde::prefixes where the inner item is already in scope.
- Both PRs ran the warning gate’s scoped check (
[0.3.135] - 2026-05-15
Added
- [Cloud][Reality] Cloud-mode Resilience dashboard — end-to-end live state from hosted-mock deployments (#468)
- Phase 1 (#517) — cloud scaffold: registry exposes
/api/v1/hosted-mocks/{deployment_id}/resilience/{circuit-breakers, bulkheads, summary}and POST reset endpoints. UI’sResiliencePagebranches onisCloudMode()and calls the registry instead of the never-mounted local/api/resilience/*routes. Originally workspace-scoped; #522 corrected the scope (see below) — circuit-breaker / bulkhead state lives in a specific mockforge process so it must be deployment-scoped. - Phase 2 (#518) — middleware in the runtime: new
mockforge_chaos::resilience_middlewareaxum layer wires every HTTP request through the existingCircuitBreakerManager+BulkheadManager. Per-endpoint circuit breaker keyed by"{METHOD} {path}"; bulkhead keyed on a configurable service string (defaults to"http"); only 5xx counts as a breaker failure (4xx stays out of the per-endpoint failure budget). Bulkhead saturation returns 503 +Retry-After: 1without recording on the breaker.mockforge servenow layers the middleware on the HTTP app and mountsmockforge_chaos::create_resilience_routeron the admin port;default_resilience_state()returns a(MiddlewareState, ResilienceApiState)pair backed by the sameArc<...Manager>instances so the dashboard reflects what the middleware records. - CLI flags wired (#519): existing
--circuit-breaker/--bulkheadflags (and the nine threshold/limit knobs that come with them) actually turn the middleware on. Newresilience_state_from_configs(circuit, bulkhead)builder takesOption<CircuitBreakerConfig>/Option<BulkheadConfig>overrides;serve.rsthreads the CLI values into bothChaosConfig(so/api/chaos/*reports identical settings) and the resilience state. When both flags are unset, behaviour is identical to before — middleware short-circuits per-request, effectively free. - Phase 3 (#522) — runtime proxy + admin enable: registry now
reqwest-proxies/api/v1/hosted-mocks/{deployment_id}/resilience/*over Fly 6PN tohttp://{HostedMock::fly_app_name}.internal:9080/api/resilience/*(3-second timeout, fail-fast).runtime_stateis now"live" | "unreachable"— dropped"pending"because every non-live state is some form of unreachable with a real proxy in place. The orchestrator now injectsMOCKFORGE_ADMIN_ENABLED=trueon every Fly deploy (bothdeploy_to_flyioandredeploy_to_flyio); without that, the admin server wouldn’t start in cloud and the proxy would have returnedunreachablefor every deployment that ever existed. UI gains a deployment selector: auto-selects the first active deployment in the org, shows a<select>only when >1 exists, renders a helpful empty state when none.
- Phase 1 (#517) — cloud scaffold: registry exposes
- [Cloud] Cloud-mode Virtual Backends — consistency lifecycle presets (#516, #461)
- Registry endpoints for managing virtual-backend lifecycle (provisioning, eventual consistency, primary/secondary fan-out) with cloud-side persistence and audit. UI lifecycle picker now drives cloud-mode workspaces without falling back to the local-only surface.
Fixed
- [Security]
lettre0.11.21 → 0.11.22 to clear RUSTSEC-2026-0141 (critical, published 2026-05-14) (#521)- Patch-level bump in Cargo.lock; no API surface change.
- Drive-by clippy cleanup in
mockforge-registryto unblock the warning ratchet on the lettre commit.
[0.3.134] - 2026-05-14
Added
- [Reality] Server-side HTTP connection lifecycle gauges (#79 round 6, Srikanth’s “how many connections are opened at a time”)
CountingMakeService(inmockforge-http) now wraps each per-connection service in aTrackedServicewhoseDropimpl records the close. The pair ofrecord_accept/record_closeincrements give an exact liveconnections_opengauge plus cumulativeconnections_total_opened/connections_total_closedcounters inmockforge_foundation::rate_counters.- Works for both plain HTTP (
axum::serve) and HTTPS (axum_server::bind_rustls) paths — only successfully accepted connections (post-TLS-handshake for HTTPS) are counted. Arecord_closealways fires perrecord_accept, even for make-service errors, so the gauge stays balanced. - When
MOCKFORGE_HTTP_LOG_CONN=1is set, each connection close emits an INFO log line under targetmockforge_http::conn_diagwithduration_msandrequests(number served before close). Combined with the existing per-requesthttp_conn_diagline, this tells you with certainty whether MockForge closed the socket after 1 request (versus the peer closing) — the exact missing piece for diagnosing Srikanth’s FIN-from-server PCAP. - Exposed in the admin/UI
SystemInforesponse:connections_open,connections_total_opened,connections_total_closed,peak_connections_open. Surfaced in the TUI dashboard’s stats panel asConns Open: N (peak M) Opened: X Closed: Y— a live multiplexing / churn indicator.
- [Reality] Bench client summary always shows connection-open counts when k6 made TCP sockets (#79 round 6, Srikanth’s “open connection on the client”)
- Previously the
Connections opened,TCP connect avg/max, andTLS handshake avg/maxlines only printed in--cpsmode. Now they print wheneverhttp_req_connecting.count > 0, so non---cpsruns can also see distinct connections opened vs request count — i.e. whether the client actually pooled connections. - New
Peak concurrent VUsline surfacesvus_maxas the upper bound on simultaneously-open client connections, paired with the server-sideConns Opengauge for cross-checking.
- Previously the
[0.3.133] - 2026-05-13
Fixed
- [Reality]
mockforge bench --rps Nproduced 0 requests under the defaultramp-upscenario (#79 follow-up, Srikanth’s 5th-round reply)- Root cause: the k6 script template derived
preAllocatedVUs/maxVUs/durationfor theconstant-arrival-rateexecutor from the last stage of the chosen scenario. The default scenario isramp-up, whose last stage is the ramp-DOWN withtarget: 0— sopreAllocatedVUs: 0and the bench completed with zero requests. Without--rps, the bench usedramping-vuswhich honors all stages, so the bug was specific to--rps. - Fix: when
target_rpsis set, the script now uses the configured--vusdirectly forpreAllocatedVUs/maxVUsand the full--durationfor the executor duration, ignoring scenario stages (which don’t apply to open-model load anyway). - Regression test added in
mockforge_bench::k6_gen::tests::test_rps_with_ramp_up_uses_full_vu_pool_and_durationthat asserts the generated script containspreAllocatedVUs: 100andduration: '600s'for a 100-VU / 600s /--rps 100invocation with the default ramp-up scenario.
- Root cause: the k6 script template derived
Added
- [Reality]
mockforge bench --cpsnow reports connections-per-second in the end-of-run summary (#79 follow-up, Srikanth’s 5th-round reply)- With
--cps(which sets k6’snoConnectionReuse: true), every request opens a fresh TCP/TLS connection, so connections/sec equals request rate. Previously the summary only printedRPS:; users running CPS-stress benches had to read it from k6’s raw output. - The terminal summary now prints
CPS:,Total Connections:, and — when k6 has samples — TCP-connect and TLS-handshakeavg/maxtimings.K6Resultsexposestcp_connect_*/tls_handshake_*fields so SDK consumers can read the same numbers programmatically.
- With
- [Reality] Opt-in
MOCKFORGE_HTTP_LOG_CONN=1env var emits per-request HTTP-version / Connection-header diagnostic log (#79 follow-up, Srikanth’s 5th-round reply)- Srikanth’s PCAP showed HTTP/1.1 requests arriving at MockForge with no
Connectionheader but MockForge sending FIN after each response. The only way to confirm what MockForge actually sees on the wire is to log the version + headers from hyper’s view. New middleware inmockforge_http::middleware::conn_diagnosticsemits one INFO log line per request withmethod,path,version,req_connection,req_keep_alive,req_host,peer,resp_status,resp_connection,resp_keep_alive, and aclose_decisionfield summarizing the keep-alive outcome (e.g.keep-alive (HTTP/1.1 default — no Connection: close)). - Disabled by default — the log is too noisy for normal operation. Truthy values:
1,true,yes,on.
- Srikanth’s PCAP showed HTTP/1.1 requests arriving at MockForge with no
[0.3.132] - 2026-05-12
Added
- [Reality] Surface every chaos fault category in
/api/chaos/stats,/metrics, and the TUI Chaos screen (#79 follow-up, Srikanth’s 4th-round reply)connection_errorfault is now recorded at the HTTP layer whenconnection_error_kind: http_503(the default). Previously only TCP-level kinds (TcpReset/TcpClose) showed up in counters, so configs that enabledconnection_errors: trueproduced 503s but no stats — exactly Srikanth’s “no connection_error in TUI” observation.- New
jitterfault counter +mockforge_chaos_jitter_mshistogram: jitter is reported separately from total injected latency so users can see jitter activity even when the base delay is zero. - New
bandwidth_throttlefault counter +mockforge_chaos_bandwidth_throttle_ms{direction}histogram: counts how oftenbandwidth_limit_bpsactually delayed a transfer and accumulates the artificial wait time, keyed byrequestvsresponsedirection. ChaosStatsSnapshotadds:latency_avg_ms_by_endpoint,jitter_samples_by_endpoint,jitter_avg_ms_by_endpoint,bandwidth_throttle_samples_by_direction,bandwidth_throttle_total_ms.- TUI Chaos
Fault Statspanel now shows mean latency, jitter sample counts + mean offset, bandwidth-throttle activity by direction, and total throttle delay.
- [Reality] Bench client surfaces server-injected chaos signals (#79 item 7)
- HTTP chaos middleware stamps three response headers when faults fire:
X-Mockforge-Injected-Latency-Ms,X-Mockforge-Injected-Jitter-Ms,X-Mockforge-Fault(e.g.partial_response). - The k6 script generated by
mockforge benchreads those headers into custom trends (mockforge_server_injected_latency_ms,mockforge_server_injected_jitter_ms) and a counter (mockforge_server_fault_total). End-of-run bench summary now prints aServer-Injected (chaos)block alongside the existing client-observed latency.
- HTTP chaos middleware stamps three response headers when faults fire:
- [Reality] New bench CLI flags
--rpsand--cps(#79 item 8)--rps N: switches the generated k6 script from the legacyramping-vusexecutor toconstant-arrival-rateatNrequests/sec, with--vusbecoming the pre-allocated VU pool. The per-iterationsleep(1)that capped throughput at ~1 req/VU/sec is dropped when--rpsis set. Use this to drive enough traffic to exercise rate-limit / connection-limit chaos that Srikanth couldn’t trigger at the legacy default 2 RPS.--cps: setsnoConnectionReuse: trueon every request so each one opens a fresh TCP/TLS connection — useful for hitting connection-limit thresholds and TCP-level fault injection.
- [Reality] Opt-in
MOCKFORGE_HTTP_KEEPALIVE_HINT=1advertisesConnection: keep-alive+Keep-Alive: timeout=N, max=Mon every response (#79 item 2 workaround)- For proxies whose upstream pool decisions read those response headers (F5/Avi/HAProxy/nginx in some configs). Won’t undo a downstream
Connection: close. Documented in the issue thread as a best-effort signal; the actual fix for the FIN/RST pattern Srikanth observed is upstream HTTP/1.1 negotiation (proxy_http_version 1.1for nginx, equivalent on other proxies).
- For proxies whose upstream pool decisions read those response headers (F5/Avi/HAProxy/nginx in some configs). Won’t undo a downstream
[0.3.131] - 2026-05-10
Added
- [Reality] TUI Chaos screen now surfaces live fault-injection counts (#79 follow-up)
- New
mockforge_chaos::metrics::ChaosStatsSnapshot— JSON-serializable view ofCHAOS_METRICSkeyed byfault_type → endpoint → count, plus per-type totals, grand total, rate-limit violations, and latency injection sample counts. - New
GET /api/chaos/stats(chaos crate) andGET /__mockforge/chaos/stats(admin passthrough) endpoints expose the snapshot as JSON. ChaosScreenadds a Fault Stats panel below Settings showing total faults, per-type counts (sorted desc by frequency), rate-limit violations, and latency injection samples. Best-effort: older servers without the new endpoint fall back to the existing 2-panel layout, so client/server version mismatches don’t break the screen.
- New
[0.3.130] - 2026-05-10
Fixed
- [Reality]
mockforge_chaos_*counters silently absent from/metrics(#79 follow-up, Srikanth’s 3rd-round reply)- Root cause: chaos counters register against
prometheus::default_registry()(viaregister_counter_vec!), but the/metricsexporter gathered only from a separate localMetricsRegistrycreated inmockforge-observability. The two registries were disjoint, so even whenrecord_fault(...)fired (wired in 0.3.128), nothing appeared at/metrics— Srikanth ran a bench that produced 80 failures andcurl /metrics | grep mockforge_chaos_returned empty. - Fix:
metrics_handlernow extends the local registry’s metric families withprometheus::default_registry().gather()before encoding, so both registries surface in the same response. No format change for clients — chaos metrics now appear alongside protocol metrics.
- Root cause: chaos counters register against
- [Reality] Chaos sub-configs (
LatencyInjectionConfig,RateLimitingConfig,NetworkShapingConfig) failed YAML parse if any field was omitted (#79 follow-up)- All three structs lacked
Defaultand#[serde(default)], so atraffic_shaping:block missingmax_connectionswould fail the whole config load. Srikanth hit this and worked around it manually. Adding the derives makes partial YAML parse cleanly with sensible zero defaults.
- All three structs lacked
[0.3.129] - 2026-05-09
Added
- [Reality] YAML config now exposes every chaos
fault_injectionfield (#79 follow-up)mockforge_core::config::FaultConfiggains the fields it was missing relative tomockforge_chaos::config::FaultInjectionConfig:connection_error_kind(http_503/tcp_reset/tcp_close),partial_responses+partial_response_probability,payload_corruption+payload_corruption_probability,corruption_type(none/random_bytes/truncate/bit_flip),error_pattern(Burst / Random / Sequential),mockai_enabled, andrequest_matcher(source IPs, headers, body-size bounds, chunked-only). All new fields use#[serde(default)]so existing chaos.yaml files continue to parse unchanged.- The bridge in
serve.rs(fault_config_to_chaos) now maps every field through to the chaos crate’s runtimeFaultInjectionConfig. Previously--config chaos.yamlset them to defaults silently and operators had to use thePUT /api/chaos/config/faultsREST API to configure them. - Tests cover round-trip parsing of the full YAML shape from the issue-#79 reply, backward-compat parsing of legacy YAML without the new fields, snake_case enum encoding, and end-to-end bridge preservation of every new field.
[0.3.128] - 2026-05-09
Fixed
- [DevX] k6 metric-name validation failure on deeply nested OpenAPI specs (Microsoft Graph etc.) (#436, #79)
- Root cause:
operationIds likedrives.drive.items.driveItem.workbook.worksheets.workbookWorksheet.charts.workbookChart.axes.categoryAxis.format.line.clear, after dot-to-underscore sanitization plus_latency/_errorssuffix, exceeded k6’s 128-char metric-name cap.validate_scriptcorrectly rejected the script before k6 ran. - New
K6ScriptGenerator::sanitize_k6_metric_namecaps the base at 112 chars (128 − 16 for_step99_latency) and appends an 8-hex-char hash of the original name when truncating, so distinct long names produce distinct metric names. JS variable identifiers keep the full readable form; only the metric string is truncated. - Wired into both render paths:
k6_script.hbs(per-operation) andk6_crud_flow.hbs. Tests cover passthrough, truncation, prefix-collision uniqueness, the “starts with letter or _” rule after truncation, and end-to-end script validation on a microsoft-graph-style operationId.
- Root cause:
- [Reality] Chaos prometheus counters were registered but never incremented (#436, #79)
mockforge_chaos_faults_total{fault_type, endpoint},mockforge_chaos_latency_ms, andmockforge_chaos_rate_limit_violations_totalall existed in the registry, but no caller invoked the correspondingrecord_*methods./metricsreported zero faults regardless of how many were actually firing — masking the effect of configured chaos rules from operators.- Wired
record_fault(...)at every fault decision point in the HTTP middleware (http_error,timeout,rate_limit,connection_limit,packet_loss,partial_response,payload_corruption) and the TCP chaos listener (tcp_reset,tcp_close). Latency injection also now records to the histogram viarecord_latency. - The TUI Chaos screen still renders config-only; surfacing these counters as a stats panel is tracked as a follow-up.
[0.3.127] - 2026-05-08
Fixed
- [Reality] TPS / RPS200 dashboard counters stuck at 0 under load (#351, #79)
- Root cause:
record_response()lives insidecollect_http_metrics, but the middleware was exported frommockforge-httpand never.layer()d onto the production router built inserve.rs. CPS kept ticking becauseCountingMakeServicewraps the make-service at a different layer. - Layer
collect_http_metricsas the outermost wrapper onhttp_appso every response — including chaos-mutated ones — bumps the rate counters the dashboard sampler reads. - Regression test pins the actual counter delta on a 2xx response, not just the response status.
- Root cause:
Added
- [DevX]
bench-chunkedaccepts--base-path, humantime--duration,--validate-requests,--export-requests(#352, #79)--base-path <PATH>prepends to every spec-derived operation path before URL construction. CLI > spec.servers > none, matchingmockforge bench. No-op without--spec.-d 600s/--duration <DURATION>switched from bare seconds (u64) to humantime parsing —30s,5m,1h, or bare seconds all parse.--validate-requests(OpenAPI request validation) and--export-requests(per-request JSON export) now mirror the same flags onmockforge bench.- The full Microsoft Graph invocation reported in #79 now parses end-to-end:
mockforge bench-chunked --chunk-size-bytes 4096 --total-size-bytes 10485760 \ --chunk-interval-ms 50 --spec microsoft-graph.yaml \ --target https://192.168.2.86 --base-path /v1.0 \ --validate-requests --export-requests --insecure -d 600s
- [DevX] Supervisor wrappers for unattended runs under heavy traffic (#350, #79)
deploy/systemd/mockforge.service— systemd unit withRestart=always, resource limits, and hardening directives. Documents the install dance and the two knobs (MemoryMax,LimitNOFILE) most likely to matter at high concurrency.deploy/scripts/run-forever.sh— bash supervisor that restarts the binary after any non-clean exit. Forwards SIGINT/SIGTERM so Ctrl-C stops cleanly. Useful on macOS / non-systemd hosts and ad-hoc bench rigs.deploy/systemd/README.mdpicks between them.
[0.3.126] - 2026-05-03
Added
- [Observability] TPS / RPS200 / CPS rate metrics in TUI dashboard + persistent CSV (#326, #79)
- TPS — successful (200..=399) responses per second
- RPS200 — 200-OK responses per second
- CPS — accepted TCP connections per second (works for both plain HTTP via
axum::serveand HTTPS viaaxum_server::Server::serve— make-service wrapper instead of listener wrapper) - Each shown current + lifetime peak in the Request Stats panel
- Three new columns appended to the
MOCKFORGE_METRICS_LOG_FILECSV:tps,rps_200,cps. Old positional CSV parsers keep working - New
mockforge-foundation::rate_countersmodule hosts the global atomic counters
- [CLI]
--https-portfor nginx-style dual HTTP+HTTPS listeners (#327, #79)- When set, the existing
--http-portlistener stays plain HTTP and a parallel TLS listener spins up on--https-port, sharing the same router and admin UI - Example:
mockforge serve --http-port 80 --https-port 443 --tls-cert server.pem --tls-key key.pem --https-portrequires--tls-cert+--tls-keyand must differ from--http-port
- When set, the existing
- [Reality]
bench-chunked --specto drive the native chunked bench from POST/PUT/PATCH operations in an OpenAPI spec (#328, #79)- When
--specis set,--targetbecomes the base URL and the bench iterates each matching operation sequentially, each running for--duration --operation-id <id>narrows to a single op
- When
- [Reality]
bench-chunkednow captures and prints up to five non-2xx response samples — status,Serverheader, first 256 bytes of body (#329, #79)- Critical for diagnosing the “503 from bench, 200 in MockForge log” pattern (almost always an upstream proxy timing out on a slow chunked upload)
- Hint output for 5xx spells out the proxy-timeout math: each request takes >=
(total_size_bytes / chunk_size_bytes) * chunk_interval_msms
Changed
- [Reality]
bench-chunkedCLI help:--total-size-bytesis now explicitly documented as per-request (not total over the run), with the chunks-per-request formula in the help text. Resolves repeat user confusion (#328, #79)
[0.3.125] - 2026-05-02
Added
- [Chaos] Per-request fault matchers (#306, #79):
request_matcher.source_ips— CIDR or bare IP allowlistrequest_matcher.headers— case-insensitive name + optional exact valuerequest_matcher.min_body_size_bytes— only requests with body ≥ Nrequest_matcher.chunked_only— onlyTransfer-Encoding: chunkedrequests- AND across fields, OR within a list. Empty matcher matches everything (back-compat)
- Applies to all five fault paths: HTTP errors, timeouts, partial responses, payload corruption, connection errors
- [Chaos] TCP-level connection errors via
ChaosTcpListener(#306, #79):connection_error_kind: tcp_reset— TCP RST at accept time (SO_LINGER=0then drop). Clients seeECONNRESETconnection_error_kind: tcp_close— TCP FIN at accept time. Clients see EOF before any HTTP responseconnection_error_kind: http_503(default) — application-layer 503 on a healthy connection (back-compat)
- [Reality]
mockforge bench-chunked— native Rust chunked-encoding traffic generator (#306, #79)- Bypasses k6 entirely. Each worker streams body via
reqwest::Body::wrap_stream, no Content-Length, guaranteed wire chunking - Supports
--concurrency,--duration,--chunk-size-bytes,--total-size-bytes,--chunk-interval-ms,--header,--insecure
- Bypasses k6 entirely. Each worker streams body via
- [TUI] Peak metrics tracked alongside current values in the dashboard (#306, #79)
- CPU, memory, error-rate now show
current (peak X)
- CPU, memory, error-rate now show
- [Observability] Persistent metrics CSV via
MOCKFORGE_METRICS_LOG_FILEenv var (#306, #79)- 10-second sampling,
timestamp,cpu_pct,mem_mb,total_reqs,err_rateper row - Survives restarts; charts in any spreadsheet, Grafana, or dashboarding tool
- 10-second sampling,
Changed
- [Chaos]
timeout_errors: truenow actuallytokio::sleep(timeout_ms)then returns 504 Gateway Timeout, applied uniformly to chunked and non-chunked. Previously this flag was incorrectly mapped to body truncation (#306, #79) - [Chaos]
partial_responsenow distinguishes chunked vs non-chunked truncation (#306, #79):- Non-chunked: truncates body but keeps original
Content-Lengthheader — clients see unexpected EOF - Chunked: truncates before terminating chunk — no
0\r\n\r\n, real protocol violation
- Non-chunked: truncates body but keeps original
- [Reality] k6 templates:
bench --chunked-request-bodiesaddsTransfer-Encoding: chunkedheader (best-effort — k6/Go’snet/httpmay still send Content-Length based on body type)
Notes
- Adds new public fields to
FaultInjectionConfig,K6Config, andK6ScriptTemplateData. External callers building these via struct literal (without..Default::default()) need a trivial update.
[0.3.124] - 2026-04-30
Fixed
- [Core]
cargo test --doc -p mockforge-corenow passes cleanly — stale doctests against types that moved out during the openapi/foundation extractions are markedignore(#285)- Validation release: closes out the chronic CI red on every release tag from v0.3.117 onward
[0.3.123] - 2026-04-30
Fixed
- [UI]
test_static_assets_content_lengthskips its size assertion when no Vite build is present, sorelease.yml(which doesn’t runpnpm) stops failing (#284)- Validation release for the chronic release-CI red
[0.3.122] - 2026-04-30
Fixed
- [UI]
admin_ui_buildtests skip cleanly whenui/dist/index.htmlis missing (#283)- Validation release for the chronic release-CI red
[0.3.121] - 2026-04-29
Fixed
- [Registry] Drop duplicate migration that blocked all sqlite migrations (#282)
_000010_user_notification_and_preferences.sqlwas a stale orphan (byte-identical to_000011_) that violated_sqlx_migrations.UNIQUE(version)against the legitimate_000010_federation_scenario_activations.sql- 26 sqlite tests in
mockforge-registry-corewere failing on every release until this landed
[0.3.120] - 2026-04-29
Fixed
- [HTTP] Mocks created via
POST /__mockforge/api/mocksnow honorMOCKFORGE_RESPONSE_TEMPLATE_EXPAND(#281)- Previously, OpenAPI-loaded route handlers expanded
{{faker.email}},{{uuid}},{{randInt …}}, etc., but mocks registered through the management API shipped the literal template strings to clients - The Node.js SDK and any test setup that creates mocks programmatically were affected
- Previously, OpenAPI-loaded route handlers expanded
- [CI]
release.yml’s test step now pre-buildsmockforgewith--features all-protocolsso e2e tests have a binary that actually starts the WebSocket and gRPC listeners (#281) - [Test]
test_data_protocol_generationno longer asserts that a JSON-string field magically becomes a JSON number after in-string template expansion (it doesn’t, and never has) (#281)
[0.3.119] - 2026-04-29
Added
- [HTTP]
--no-rate-limitCLI flag andMOCKFORGE_RATE_LIMIT_ENABLED=falseenv var to fully disable the per-IP HTTP rate limiter (#280, #79)MOCKFORGE_RATE_LIMIT_DISABLED=trueis a documented alias- Reported by @srikr while load-testing the api.github.com spec — sustained load was hitting the default 1000 RPM / 2000 burst limits and returning 429s with
Retry-After: 60 - Workaround prior to this release was setting
MOCKFORGE_RATE_LIMIT_RPM=100000000/MOCKFORGE_RATE_LIMIT_BURST=100000000; that still works and is unchanged
[0.3.104] – [0.3.118] — 2026-03-29 to 2026-04-23
These releases predate the chore(release): bump workspace to X.Y.Z
commit convention introduced in 0.3.119, so there’s no single commit
to lift detail from per version. Per-release notes are auto-generated
on each GitHub release page:
The crates are all published and resolvable on crates.io
(cargo search mockforge-cli lists every version).
[0.3.103] - 2026-03-28
Added
- [Reality] Field-level schema validation errors in conformance failure details (#79)
- Violations now show specific field path, violation type, expected/actual values
- Uses
jsonschema::validate()instead ofis_valid()for detailed error reporting - Displayed in terminal output and
conformance-failure-details.json
- [Reality] HAR-to-YAML generator:
mockforge har-to-conformance --har file.har(#79)- Converts browser HAR captures to custom compliance YAML
- Auto-detects base URL, filters static assets, extracts response headers and JSON body field types
- Output compatible with
--conformance-custom
- [Reality] Multi-target conformance:
--conformance+--targets-filenow works (#79)- Runs conformance tests against each target sequentially using native executor
- Per-target reports in
output/target_N/conformance-report.json - Combined summary in
multi-target-conformance-summary.json
[0.3.102] - 2026-03-28
Fixed
- [Reality] Add
summary.jsonto spec-driven conformancehandleSummary(#79)- Custom conformance tests (via
--conformance-custom) now generatesummary.json
- Custom conformance tests (via
[0.3.101] - 2026-03-26
Fixed
- [Reality] Skip automatic Authorization (Basic/Bearer) headers when Cookie header is provided via
--conformance-header(#79)- Users managing session-based auth no longer get conflicting Basic Auth headers
- [Reality] Add
summary.jsonoutput to reference conformance generator’shandleSummary(#79)
[0.3.100] - 2026-03-24
Fixed
- [Core] Prevent panic on conflicting parameter names in OpenAPI routes (#79)
- Routes with different param names at the same position (e.g.,
{attestation_id}vs{subject_digest}) now canonicalize to first-registered names - Fixes crashes when loading GitHub, Xero, and similar real-world specs
- Applied to all 4 router builders
- Routes with different param names at the same position (e.g.,
- [Core] Swagger 2.0
formDataparameters now properly converted to OpenAPI 3requestBody(#79)application/x-www-form-urlencodedfor regular fields,multipart/form-datafor file uploads
[0.3.99] - 2026-03-23
Fixed
- [Core] Fix param name conflict panic — same as v0.3.100 (v0.3.99 publish was overwritten by concurrent process)
[0.3.98] - 2026-03-22
Fixed
- [Reality] Fix zero stats in multi-target summary — removed deprecated
--summary-exportflag; rely solely onhandleSummary()for writingsummary.json(#79) - [Reality] Fix failed_requests metric reading
.fails(success count) instead of.passes(failure count) from k6 Rate metric (#79) - [Reality] Fix k6 script path not found when CWD is set to output dir — now uses absolute script path (#79)
- [Reality] Fix
--summary-exportabsolute path for CWD mismatch (#79) - [Reality] Fix k6 API server port conflict in multi-target mode — each parallel instance gets a unique port (#79)
- [Reality] Treat k6 exit code 99 (thresholds crossed) as warning, still parse results (#79)
- [Reality] Warn when
--conformanceis used with--targets-file(not yet supported) (#79) - [Core] Downgrade contract diff
$refresolver warnings from WARN to DEBUG (#79)
Added
- [Reality] Total elapsed time in multi-target summary output and
aggregated_summary.json(#79) - [Reality]
all_targets.csvfile with per-target metrics for easy parsing (#79) - [UI] Host header column in admin dashboard Recent Logs (#79)
- [TUI] Client IP and Host columns in both Dashboard and Logs screens (#79)
[0.3.91] - 2026-03-17
Fixed
- [HTTP] Fix double-slash in conformance URLs when
--targethas trailing slash (#79) - [HTTP] Apply OData rewrite layer to TLS/HTTPS server path (was only on non-TLS) (#79)
- [CLI] Downgrade non-actionable WARN messages to DEBUG/INFO (MockAI key, proto dir, auth backend, JWT secret) (#79)
[0.3.90] - 2026-03-16
Added
- [Core] OData function call path support via URI rewrite layer (#79)
- Handles paths like
(period='{period}')for Microsoft Graph compatibility - Mock responses generated for OData function endpoints
- Handles paths like
[0.3.89] - 2026-03-15
Fixed
- [Core] Gracefully skip OData function call paths in OpenAPI specs instead of failing (#79)
[0.3.88] - 2026-03-14
Added
- [Reality]
--conformance-delayflag to add delay between conformance requests (#79) - [Reality] k6 output logging to file for debugging (#79)
- [Reality] 429 rate-limit detection and error clarity in conformance output (#79)
Fixed
- [Cloud] End-to-end deployment pipeline, storage fallback, remaining console errors
- [CLI] Load OpenAPI spec from
MOCKFORGE_CONFIGenv var for cloud deployments - [Cloud] Auto-detect Fly.io registry images for cross-app pulls
[0.3.85] - 2026-03-12
Added
- [Reality] Native conformance executor with API, UI dashboard, and SDK integration (#79)
- [Reality] Conformance report UX improvements and custom test authoring (#79)
- [Reality] Full request/response detail capture for conformance failures (#79)
- [Reality] OWASP API Top 10 coverage mapping in conformance reports (#79)
Fixed
- [Reality] Eliminate misleading error rate in conformance output (#79)
- [Reality] Deduplicate native executor checks, write failure details file (#79)
- [Reality] Custom conformance checks now emit failure details (#79)
- [Core/Bench] Resolve 3 conformance test failures (#79)
- [UI/Bench] Fix empty routes and add endpoint details to conformance (#79)
[0.3.80] - 2026-03-10
Fixed
- [Analytics] Make migrations idempotent with
IF NOT EXISTSguards (#79) - [TUI] Add g/G and PgUp/PgDn to routes screen status hint (#79)
- [UI/Bench] Fix routes proxy, improve OWASP coverage reporting (#79)
Added
- [Cloud] Cloud mode support, auth fixes, runtime error hardening
- [Cloud] Deployment deletion CLI command and background cleanup worker
- [HTTP/UI] API explorer for hosted mock deployments
Changed
- [Refactor] Architectural overhaul — 10 workstreams across core, UI, CI
- [Perf] Replace linear route scan with matchit trie-based matching
- [Refactor] Extract chaos modules from core into mockforge-chaos
- [Refactor] Split config.rs and openapi/response.rs into submodules
- [Refactor] Restrict 8 internal modules to pub(crate), add 35 GraphQL tests
- [Refactor] Implement MockProtocolServer trait for all 10 protocols
[0.3.76] - 2026-03-08
Fixed
- [UI/TUI] Implement 11 missing admin API endpoints that caused “40 Errs” in the TUI dashboard (#79)
- The TUI polls
/__mockforge/chaos,/__mockforge/recorder/status,/__mockforge/world-state,/__mockforge/federation/peers, and/__mockforge/vbr/statusevery 2-30 seconds - These endpoints previously returned 404, incrementing the TUI error counter ~22 times/minute
- All 5 GET endpoints now return valid JSON with real data from live subsystem instances
- 6 POST/DELETE mutation endpoints (
chaos/toggle,recorder/start,recorder/stop,chaos/scenarios/{name}start/stop) are auth-gated via RBAC - Added 5 new handler modules:
chaos_api,recorder_api,world_state_proxy,federation_api,vbr_api - 26 E2E tests verified against a real running server (GET responses, auth rejection, auth acceptance, state mutations)
- The TUI polls
- [CLI] Admin server now creates real subsystem instances instead of returning empty defaults (#79)
- Recorder: created from
--recorder-dbconfig when--recorderflag is passed - VBR engine: created with in-memory storage backend (lightweight, no disk side-effects)
- Federation: empty instance created so the TUI shows a valid (but empty) federation state
- Recorder: created from
- [TUI] Error counter capped at 999 to prevent unbounded growth (#79)
- [CLI] Fix pre-existing
clippy::print_literalwarning in cloud commands
Added
- [Federation]
Federation::empty()constructor for creating a default empty federation instance - [CLI]
mockforge-federationadded as a dependency for admin server integration
[0.3.73] - 2026-03-05
Fixed
- [UI] Fix
cargo publishfailure formockforge-uicaused bybuild.rsmodifying source directory- Removed code that copied
pwa-manifest.jsonandsw.jsintoui/dist/during build (violates cargo’s source-dir-immutability rule) serve_service_workernow readssw.jsfromui/public/(same pattern asserve_manifest)- Added
sw.jsto the crate’sincludelist so it’s packaged correctly
- Removed code that copied
- [Core] Mock server now supports
X-Mockforge-Response-Statusheader to return non-default status codes (#79)- Conformance checks for
response:404andresponse:400previously always failed because the server returned the first declared status (usually 200) - New
has_response_for_status()validates the requested code exists in the spec before overriding - Both OpenAPI handler paths extract and pass the header through
- Conformance checks for
- [Core] Response generation no longer replaces object-typed properties with string examples (#79)
- When a property schema declares
type: object, the fallback now preserves an empty{}instead of generating a name-based string like"example config" - Fixes
response:schema:validationfailures where JSON schema validation rejected string values for object properties - Added
is_object_typed_property()helper for type-aware fallback decisions
- When a property schema declares
- [Data]
generate_by_type("object")now returns{}instead of"unknown_type_object"(#79)- Also added
"array"handler returning[]
- Also added
Changed
- [Reality] Spec-driven conformance generator now sends
X-Mockforge-Response-Statusheader forresponse:400andresponse:404checks (#79)- Tells the mock server which status code to return, enabling accurate status-code conformance testing
[0.3.72] - 2026-03-04
Fixed
- [UI]
mockforge serve --adminno longer panics when no production auth is configured (#79)validate_auth_config_on_startup()now logs a warning instead of returning an error- Auto-generated JWT secret fallback so the admin UI works out of the box
- Default users are seeded even without
ENVIRONMENT=development, so login works immediately
- [Reality] Fix duplicate session ID in conformance Cookie headers (#79)
- Removed invalid
noCookies: truefrom k6 options (not a real k6 option; k6 silently ignores it) - Added
http.cookieJar().clear(BASE_URL)before and after each request when custom Cookie headers are present - Prevents k6’s internal cookie jar from re-sending server
Set-Cookievalues alongside custom headers - Applied to both reference-mode (
generator.rs) and spec-driven (spec_driven.rs) generators
- Removed invalid
- [Reality] Fix missing single-quote escaping in spec-driven
format_headers()(#79)- Header values containing single quotes are now properly escaped, matching
generator.rsbehavior
- Header values containing single quotes are now properly escaped, matching
Added
- [Reality] Conformance report now shows individual failed checks with pass/fail counts (#79)
- New “Failed Checks” section after the category summary table lists each check that failed
- When not using
--conformance-all-operations, prints a tip suggesting it for endpoint-level detail
[0.3.70] - 2026-02-27
Fixed
- [Reality] Remove dead
CUSTOM_HEADERSJS const from conformance generators (#79)- Custom header values are now inlined directly into each request instead of referencing an unused JS constant
- Eliminates confusing dead code in generated k6 scripts
- [Reality] Add
noCookies: trueto k6 options when Cookie header is in custom headers (#79)- Prevents k6’s automatic cookie jar from duplicating cookies on subsequent requests
- Fixes duplicate session ID / authentication failures reported by @srikr
- [Reality] Fix conformance report file not found after k6 execution (#79)
handleSummarynow writesconformance-report.jsonto an absolute path matching the output directory- Previously wrote to a relative path based on k6’s CWD, causing the CLI to report “Conformance report not generated”
Added
- [Reality]
--conformance-all-operationsflag for full-endpoint conformance testing (#79)- Default mode tests one representative operation per feature check (fast feature-coverage)
- New flag tests ALL operations with path-qualified check names (e.g.,
method:GET:/api/users) - Addresses user confusion about “only 5 endpoints tested”
- [Reality] Conformance coverage summary output (#79)
- After generating conformance tests, prints “Conformance: N operations analyzed, M unique checks generated”
- When using default mode with fewer checks than operations, shows tip about
--conformance-all-operations
[0.3.69] - 2026-02-24
Fixed
- [Multi] Replace 36+
assert!(true)placeholder tests with meaningful assertions across 16 files- CLI command tests (MQTT, SMTP, governance) now construct and verify command variants
- Registry server tests use compile-time type checks instead of no-op assertions
- Integration tests (voice workspace, drift GitOps, behavioral cloning, WebSocket, cross-platform sync) use proper verification patterns
- [gRPC] Add
use super::*to 13 emptytest_module_compiles()tests so they actually verify module compilation - [HTTP] Fix misleading “placeholder” doc comment on fully-implemented
get_proxy_inspecthandler
[0.3.57] - 2026-02-14
Fixed
- [Reality] Spec-driven conformance: global security requirement detection (#79)
annotate_security()now falls back tospec.security(root-level) when an operation has no operation-level security defined- APIs that only define security globally are now correctly detected
- [Reality] Spec-driven conformance: SecurityScheme type resolution (#79)
- Security schemes are now resolved from
components.securitySchemesto detect actual type (HTTP/bearer,APIKey,HTTP/basic) instead of relying on name heuristics alone - A scheme named “myAuth” that is actually an
apiKeytype is now correctly identified - Name-based heuristic retained as fallback for unresolvable schemes
- Security schemes are now resolved from
- [Reality] Spec-driven conformance: ContentNegotiation detection (#79)
ContentNegotiationfeature is now detected when a response defines multiple content types (e.g., bothapplication/jsonandapplication/xml)- Previously only worked in reference mode
- [Reality] CLI help text for
--conformance-categoriesnow includesresponse-validation(#79)
Added
- [Reality] 5 new conformance tests: ResponseValidation with schema check, global security, SecurityScheme resolution, ContentNegotiation detection, single-type negative case (#79)
[0.3.56] - 2026-02-14
Added
- [Reality] Conformance category filtering (#79)
- New
--conformance-categoriesflag to run only specific conformance categories (e.g.,--conformance-categories "parameters,security") - Case-insensitive category matching with validation against known categories
- New
- [Reality] Spec-driven conformance testing (#79)
- When
--conformance --spec my-api.jsonis provided, analyzes the user’s actual OpenAPI spec to detect which features their API exercises - Generates conformance tests against real endpoints instead of reference
/conformance/paths - Full
$refresolution with cycle detection for parameters, schemas, request bodies, and responses - Detects: parameter types, request body formats, schema types/composition/formats/constraints, response codes, security schemes
- When
- [Reality] Response schema validation (#79)
- In spec-driven mode, validates response bodies against OpenAPI response schemas
SchemaValidatorGeneratorproduces JavaScript validation expressions from OpenAPI schemas- Supports object (required fields, property types), array, string (format regex, enum, length), integer/number (range), boolean validation
- Wrapped in try-catch for resilient k6 execution
- [Reality] SARIF 2.1.0 report output (#79)
- New
--conformance-report-format sarifflag outputs conformance results in SARIF 2.1.0 format - Compatible with GitHub Code Scanning, VS Code SARIF Viewer, and CI/CD pipelines
- Maps each conformance feature to a SARIF rule with OpenAPI spec section links
- Passed features emit
level: "note", failed features emitlevel: "error"
- New
[0.3.55] - 2026-02-14
Added
- [Reality] Per-server stats in multi-target mode (#79)
K6Resultsnow parses RPS, VUs, and full latency breakdown (min/med/p90/p95/p99/max) from k6summary.jsonAggregatedMetricsincludestotal_rps,avg_rps,total_vus_max- Multi-target reporter shows per-target RPS, VUs, and full latency breakdown
aggregated_summary.jsonincludes all new metrics in both aggregated and per-target sections
- [Reality] Per-target spec support for multi-target mode (#79)
- Targets file JSON format now supports
"spec"field for per-target OpenAPI specs - Each target can use a different spec file for heterogeneous fan-out
- Example:
[{"url": "https://server1", "spec": "spec_a.json"}, {"url": "https://server2", "spec": "spec_b.json"}]
- Targets file JSON format now supports
- [Reality] OpenAPI 3.0.0 conformance testing (#79)
- New
--conformanceflag generates and runs comprehensive k6 scripts exercising 47 OpenAPI 3.0.0 features across 10 categories (Parameters, Request Bodies, Schema Types, Composition, String Formats, Constraints, Response Codes, HTTP Methods, Content Negotiation, Security) - Reports per-category pass/fail rates with colored terminal output
- Supports
--conformance-api-key,--conformance-basic-auth,--conformance-reportfor security scheme testing - Example:
mockforge bench --conformance --target http://localhost:3000
- New
[0.3.54] - 2026-02-13
Fixed
- [Reality] fix(bench): deliver CRS payloads as path injection + form-encoded body (#79)
- Added
inject_as_pathfield toSecurityPayload— URI payloads without query params (e.g., CRS 942101:POST /1234%20OR%201=1) now replace the request path viaencodeURI()so WAFs inspectREQUEST_FILENAMEinstead ofARGS - Added
form_encoded_bodyfield toSecurityPayload— body payloads from CRS tests (e.g., 942432:var=;;dd foo bar) now sent asapplication/x-www-form-urlencodedso WAFs parse form data intoARGSfor character counting - Updated
k6_script.hbsandk6_crud_flow.hbstemplates to handle both new delivery mechanisms - Replaced unreliable
startsWith('/')URI heuristic in CRUD flow template with explicitinjectAsPathflag - Expected SQLi detection: 46/46 rules (100%), up from 45/46 (97.8%)
- Added
[0.3.53] - 2026-02-13
Fixed
- [Reality] fix(bench): URL-encode URI payloads + strip form keys from body payloads (#79)
- URI security payloads now wrapped in
encodeURIComponent()for valid HTTP transport — WAFs decode before inspection (fixes 942101) - Form-encoded body payloads now have form key prefix stripped (
var=;;dd foo bar→;;dd foo bar) so WAF ARGS parsing sees the attack payload directly (fixes 942432) - Confirmed SQLi detection: 45/46 rules (97.8%), up from 43/46 (93.5%)
- URI security payloads now wrapped in
[0.3.52] - 2026-02-12
Fixed
- [Reality] fix(bench): Group multi-part WAFBench payloads + decode body payloads + fix Cookie/CookieJar conflict (#79)
- Multi-part CRS test cases (URI + headers + body) now grouped by
group_idand sent together in one HTTP request instead of being split across separate requests (fixes 942290) - Body payloads from CRS YAML files are now form-URL-decoded before injection (
%22+WAITFOR+DELAY+%27→" WAITFOR DELAY ') so WAFs see actual SQL patterns in JSON bodies (fixes 942240, 942320, 942432) - URI payloads from path-only CRS tests are now URL-decoded and stripped of leading
/artifact (fixes 942101) - Cookie header payloads no longer overridden by empty CookieJar —
secRequestOptsconditionally skipsjar: new http.CookieJar()when a security Cookie header is present (fixes 942420, 942421) - Added
groupedPayloadsarray-of-arrays in generated k6 scripts;getNextSecurityPayload()returns arrays of related payloads - Template loop applies URI/header/body parts simultaneously per request via
secPayloadGroup - Expected SQLi detection improvement: 37/46 → 44/46 (80.4% → 95.7%)
- Multi-part CRS test cases (URI + headers + body) now grouped by
[0.3.51] - 2026-02-11
Fixed
- [Reality] fix(bench): Accept all WAFBench CRS payloads without attack-pattern filter (#79)
- Removed overly strict
attack-patterncategory filter that was silently dropping valid CRS test cases - All CRS YAML test cases now loaded regardless of their
attack_typemetadata
- Removed overly strict
[0.3.50] - 2026-02-10
Fixed
- [Reality] fix(bench): Use per-request CookieJar instead of shared EMPTY_JAR (#79)
- Each HTTP request now creates its own
new http.CookieJar()instead of sharing a global empty jar - Prevents cookie cross-contamination between requests in security testing
- Each HTTP request now creates its own
[0.3.49] - 2026-02-09
Fixed
- [Reality] fix(bench): Send raw security payloads + use dedicated empty cookie jar (#79)
- Security payloads now sent as raw strings without additional encoding
- Dedicated empty CookieJar per request prevents k6’s default cookie accumulation
[0.3.48] - 2026-02-08
Fixed
- [Reality] fix(bench): Cycle security payloads per-operation + clear cookies in API2 tests (#79)
- Security payloads now cycle per-operation block (each API endpoint gets a different payload)
- Previously all operations in one VU iteration used the same payload
- OWASP API2 (Broken Auth) tests now properly clear cookies between requests
[0.3.47] - 2026-02-06
Added
- [DevX] chore: Add Claude Code setup (CLAUDE.md, agents, skills, hooks, hookify)
- Project-specific Claude Code configuration with rules, agents, and skills
- Custom skills for verification, template checking, code review, and bench review
- Hookify rules engine for behavioral guardrails
Fixed
-
[Reality] fix(bench): Security payloads now injected + cookie dedup in all templates (#79)
- Security payloads now properly injected in both k6_script.hbs and k6_crud_flow.hbs templates
- Cookie deduplication applied to all HTTP request paths in both templates
- Comprehensive test suite added for issue #79 security pipeline
-
[Registry] fix(registry): Add RBAC permission system with Display, AdminAll bypass, and PermissionChecker
- New RBAC permission model with role-based access control
- AdminAll role bypasses all permission checks
- PermissionChecker trait for consistent authorization across endpoints
[0.3.46] - 2026-01-30
Fixed
-
[Reality] fix(bench): WAFBench payloads now distributed across VUs for better coverage (#79)
- Changed payload cycling to use VU-based offset:
(__VU - 1) % payloads.length - Previously all 50 VUs started at index 0 and cycled through same sequence
- Now each VU starts at a different payload, maximizing attack coverage in shorter test runs
- With 50 VUs and 30 payloads, all payloads are tested from the start
- Changed payload cycling to use VU-based offset:
-
[Reality] fix(bench): OWASP API tests now include custom headers in all requests (#79)
- Added
CUSTOM_HEADERSto API8 verbose error test (malformed JSON body test) - Added
CUSTOM_HEADERSto API9 discovery paths test - Added
CUSTOM_HEADERSto API9 API versions test - Fixes auth failures when using
--headers "Cookie:..."with OWASP testing
- Added
[0.3.43] - 2026-01-16
Fixed
- [Reality] fix(bench): Security payloads now actually applied to requests in k6 scripts (#79)
- Updated k6_script.hbs template to call
getNextSecurityPayload()andapplySecurityPayload() - Previously, security payload functions were defined but never called in generated scripts
- Security payloads now properly injected into request bodies for POST/PUT/PATCH
- Header-based payloads now properly injected into request headers
- Updated k6_script.hbs template to call
[0.3.42] - 2026-01-15
Fixed
- [Reality] fix(bench): XSS payloads now inject into ALL string fields, not just the first one (#79)
- Removed
breakstatement fromapplySecurityPayload()loop in security_payloads.rs - Ensures WAF can detect payloads regardless of which field it scans
- Removed
- [Reality] fix(bench): Added
jar: nullto remaining OWASP HTTP calls to prevent cookie duplication (#79)- Fixed testBrokenAuth empty token test
- Fixed testMisconfiguration verbose error test
- Fixed testInventory discovery paths and API versions checks
- [CLI] fix(cli): Fixed format string compilation error in plugin_commands.rs (#79)
- Escaped all braces (
{→{{,}→}}) insideformat!macro for auth plugin template - Fixes “invalid format string: expected
}, foundr” compilation error
- Escaped all braces (
[0.3.39] - 2026-01-14
Fixed
- [Reality] fix(bench): WAFBench XSS attacks now properly injected into request body (#79)
- Removed location check from
applySecurityPayload()- ALL payloads now injected into body for POST/PUT - WAFBench payloads correctly pass location info (uri/header/body) to k6 scripts
- Header payloads include header name for proper injection into specified headers
- Removed location check from
- [Reality] fix(bench): Cookie header duplication in OWASP and security tests (#79)
- Added
jar: nullto all HTTP request params to disable k6’s automatic cookie jar - Prevents duplicate cookies when user provides Cookie header via
--headersflag - Applied to k6_script.hbs, k6_crud_flow.hbs, and OWASP generator
- Added
[0.3.38] - 2026-01-13
Fixed
- [Reality] fix(bench): pass custom headers from
--headersflag to OWASP tests (#79)- Cookie and other custom headers are now included in all OWASP request helpers
- Fixes issue where
avi-sessionid=Nonewas being sent instead of actual cookie values
- [Reality] fix(bench): WAFBench loader now handles single YAML file paths (#79)
- Previously only directories or glob patterns were supported
- Single file paths like
/path/to/941100.yamlnow work correctly
- [Reality] Verified CRS v3.3 format compatibility with full CoreRuleSet test suite
- Tested with 175 files, 1512 payloads (692 XSS, 505 SQLi, 304 Command Injection, 11 Path Traversal)
[0.3.37] - 2026-01-12
Added
- [Reality] feat(bench): add WAFBench cycle-all mode (
--wafbench-cycle-all) to test all payloads sequentially (#79) - [Reality] feat(bench): add
--owasp-iterationsparameter to control OWASP test iterations per VU (#79) - [Reality] feat(bench): OWASP tests now respect
--vusparameter for concurrent testing (#79)
Fixed
- [Reality] fix(bench): WAFBench payloads now properly injected in standard bench mode (not just CRUD flow)
- [Reality] fix(bench): OWASP APIs now use random UUIDs per request instead of static IDs for BOLA testing (#79)
- [Reality] fix(bench): OWASP auth tokens with special characters (quotes, backslashes) now properly escaped (#79)
- [Reality] fix(bench): prevent Handlebars double-escaping of pre-escaped JavaScript values
- [Reality] fix(bench): WAFBench security payloads now integrated into CRUD flow requests (#79)
- [Reality] fix(owasp): use
http.del()instead ofhttp.delete()for k6 compatibility (#79) - [Reality] fix(owasp): add
--base-pathsupport for OWASP API testing (#79) - [Reality] fix(bench): remove undefined
totalRequestCountvariable reference - [Reality] fix(bench): support CRS v3.3 WAFBench format and pass
--insecureto OWASP tests
[0.3.33] - 2026-01-10
Fixed
- [Reality] fix(bench): multiple fixes for OWASP and WAFBench testing
- Support CRS v3.3 format in WAFBench parser
- Pass
--insecureflag to OWASP tests for self-signed certificates
[0.3.31] - 2026-01-08
Fixed
- [Reality] fix(bench): fix extracted value substitution in CRUD flows
- [Reality] fix(bench): OWASP k6 configuration improvements
[0.3.30] - 2026-01-07
Added
- [Reality] feat(bench): add
merge_bodysupport for CRUD flows - merge extracted values with request body - [Reality] feat(bench): add
inject_attacksdata model for security testing in CRUD flows
[0.3.28] - 2026-01-06
Added
- [Reality] feat(bench): add nested path extraction for CRUD flows (e.g.,
results[0].id) - [Reality] feat(bench): add filter extraction for CRUD flows (e.g.,
results[?name=='test'].id)
[0.3.27] - 2026-01-05
Added
- [Reality] feat(bench): add full body extraction for CRUD flows
- [Reality] feat(bench): add key filtering for extracted values
[0.3.26] - 2026-01-04
Added
- [Reality] feat(bench): add aliased extraction for CRUD flow value chaining
- Extract values with aliases (e.g.,
id as poolId) for use in subsequent requests
- Extract values with aliases (e.g.,
[0.3.24] - 2026-01-03
Fixed
- [Reality] fix(bench): use correct variable name in CRUD flow extracted value replacement
[0.3.22] - 2026-01-02
Added
- [Reality] feat(bench): add OWASP API Security Top 10 testing mode (#79)
- Test for BOLA (API1), Broken Auth (API2), Mass Assignment (API3), Resource Consumption (API4)
- Test for Function Auth (API5), SSRF (API7), Misconfiguration (API8), Inventory (API9), Unsafe Consumption (API10)
- Configurable test categories with
--owasp-categories - Support for auth tokens with
--owasp-auth-token - SARIF and JSON report formats
Changed
- [DevX] chore: include UI dist files for publishing to crates.io
[0.3.21] - 2025-12-31
Fixed
- [DevX] fix(bench): use custom flow config and fix sequential mode path matching - enables cross-resource dependency chains
- [DevX] fix(bench): process dynamic placeholders in CRUD flow params file bodies (#79)
- chore: update benchmark baseline [skip ci]
- chore: enable publishing for previously internal crates
- chore: update benchmark baseline [skip ci]
- fix(release): disable sccache for crates.io publish
- chore: update benchmark baseline [skip ci]
- fix(release): publish all crates in dependency order
- fix(release): add mockforge-core to crates.io publish order
- chore: update benchmark baseline [skip ci]
- feat(bench): add –base-path option for API base path support (#79)
- chore: update benchmark baseline [skip ci]
- fix(collab): include SQLx query cache for crates.io installation (#79)
- chore: update benchmark baseline [skip ci]
- feat: implement optional enhancements from improvement plan
- fix: update doc tests to use rust,ignore for external dependencies
- chore: update benchmark baseline [skip ci]
- chore: add missing crates to workspace and restore path dependencies
- chore: restore path dependencies after publishing remaining v0.3.17 crates
- fix: restore all crates to workspace members list
- chore: restore path dependencies after publishing v0.3.17
- docs: update CHANGELOG for v0.3.17 release
- feat(bench): add WAFBench YAML integration for security testing
- Bump version to 0.3.17
- feat: comprehensive improvements across AMQP, MQTT, gRPC, registry server, and UI
- feat(ui): add type safety, mobile layout fixes, and search/filter to frontend
- Restore path dependencies after publishing v0.3.16
- Bump version to 0.3.16
- fix: resolve flaky tests and race conditions across test suite
- fix: replace panic-prone unwrap calls with safe error handling
- fix: resolve UUID storage format mismatch in collab crate tests
- Add multi-spec support and cross-spec dependency detection for bench command
- feat: add multi-spec support and cross-spec dependency handling to bench command
- fix: add validation to CRUD flow script generation
- fix: sanitize k6 CRUD flow metric names (#79 follow-up)
- Bump version to 0.3.13 and improve changelog
- Bump version to 0.3.12 and publish to crates.io
- Bump version to 0.3.11 and publish to crates.io
- chore: update benchmark baseline [skip ci]
- feat: add –params-file option for custom parameter values in bench
- Bump version to 0.3.10 and publish to crates.io
- chore: update benchmark baseline [skip ci]
- fix: move insecureSkipTLSVerify to global k6 options (fixes –insecure)
- chore: update benchmark baseline [skip ci]
- fix: resolve k6 bench issues with –insecure flag, textSummary, and query params
- chore: update benchmark baseline [skip ci]
- chore: bump version to 0.3.9 and update changelog
- feat: implement comprehensive mock server functionality across all crates
- chore: commit remaining version updates
- fix: enable publishing for mockforge-ui
- fix: enable publishing for mockforge-tunnel
- fix: update all 0.3.7 dependencies to 0.3.8 with path dependencies
- fix: add path dependencies for all workspace crates
- chore: update CHANGELOG date for 0.3.8
- chore: bump version to 0.3.8
- Fix cargo publish issues: add version requirements to dependencies
- chore: update benchmark baseline [skip ci]
- Apply formatting and additional code changes
- Fix compilation errors: update dependencies and adapt to API changes
- fix: remove path from mockforge-pipelines dep in mockforge-collab
- Add mockforge-sdk, mockforge-ui, mockforge-cli to workspace
- fix: add mockforge to restore function targets list
- fix: convert mockforge dev-dependencies to path dependencies
- fix: add mockforge-core to restore list and manually fix dependency
- fix: include mockforge-core in restore list
- fix: restore function now properly handles table-form dependencies without path
- fix: automatically restore dependencies at start of publish
- fix: restore all crate dependencies, not just a few
- fix: only convert dependencies for already-published crates
- fix: correct publish order - publish mockforge-data before mockforge-core
- fix: add mockforge-data as optional dependency in mockforge-core
- chore: bump version to 0.3.6 and update changelog
- chore: update benchmark baseline [skip ci]
- Fix k6 script generation and UI icon embedding issues
- chore: update benchmark baseline [skip ci]
- Add comprehensive test suite and fix build issues
- chore: update benchmark baseline [skip ci]
- docs: add comprehensive performance benchmarks documentation
- chore: update benchmark baseline [skip ci]
- fix: implement real functionality in benchmark tests and fix k8s-operator
- chore: update benchmark baseline [skip ci]
- fix: filter out ‘change’ directories from benchmark baseline parsing
- chore: update benchmark baseline [skip ci]
- chore: update benchmark baseline [skip ci]
- fix: GitHub Actions workflow cleanup and fixes (#81)
- chore: restore dependencies after publishing all crates
- fix: add mockforge-cli to workspace and add metadata to mockforge-k8s-operator
- fix: add missing crates to workspace (mockforge-sdk, mockforge-http, mockforge-ui, mockforge-k8s-operator)
- fix: add mockforge-world-state to workspace and publishing order before mockforge-http
- fix: add mockforge-route-chaos publishing step before mockforge-http
- fix: add mockforge-route-chaos to dependency targets and publishing order
- fix: add mockforge-route-chaos to workspace and publishing script
- fix: add mockforge-route-chaos to publishing order before mockforge-http
- fix: reduce keywords from 6 to 5 for mockforge-performance
- fix: reduce keywords to 5 for mockforge-performance (crates.io limit)
- fix: add mockforge-performance to publishing order before mockforge-http
- fix: add mockforge-collab to workspace members list
- fix: add mockforge-collab to workspace members
- fix: add missing README.md for mockforge-pipelines
- fix: add mockforge-pipelines to publishing order and dependency targets
- fix: add mockforge-pipelines to workspace and publishing script
- fix: add all missing crates to workspace members
- fix: handle short form dependencies when converting to path
- fix: publish mockforge-template-expansion before mockforge-core
- fix: add mockforge-template-expansion to publishing script
- fix: temporarily convert dependent crates’ dependencies to path before publishing
- fix: remove argon2 from mockforge-core during MSRV checks
- fix: exclude mockforge-collab from MSRV checks and remove patch section
- fix: use awk instead of sed for multi-line patch section insertion
- fix: use Cargo patch section to pin base64ct for MSRV
- fix: improve base64ct pinning order in MSRV workflow
- fix: use exact version constraint for base64ct in MSRV workflow
- fix: improve base64ct pinning in MSRV workflow
- fix: pin base64ct to 1.7 for MSRV compatibility
- fix: exclude mockforge-ui from MSRV checks
- fix: add abd and existant to typos config
- fix: exclude FontAwesome and all minified files from spell check
- fix: also remove sysinfo from mockforge-ui during MSRV checks
- fix: exclude elasticlunr.min.js from spell check
- fix: exclude highlight.js from spell check
- fix: disable sysinfo feature during MSRV checks
- fix: sync sysinfo to 0.37, fix resolvable typo, exclude ace.js from spell check
- fix: pin sysinfo to 0.36, fix typos, improve MSRV workaround
- fix: update MSRV to 1.80 and add GraphQL exclusion workaround
- fix: update MSRV from 1.82 to 1.75
- fix: fix GitHub Actions workflow failures
- fix: standardize dependencies and fix all test failures
- Skip CRDs in kubectl validation to avoid server connection
- Fix kubectl validation to prevent server connection attempts
- Fix kubectl validation to skip server connection
- Fix all test failures and resolve dependency conflicts
- Fix k6 metric name validation error (issue #79) (#80)
- Optimize workflows: update deprecated actions and add path filters
- Fix mockforge-smtp version constraint from 0.2.0 to 0.3.3
- Fix Docker build, k8s validation, and spell check issues
- fix: update all mockforge dependency versions to 0.3.3 in mockforge-http
- chore: fix formatting (pre-commit hooks)
- deps(deps): bump opentelemetry_sdk from 0.21.2 to 0.31.0 (#67)
- chore: update benchmark baseline [skip ci]
- deps(deps): bump opentelemetry-semantic-conventions (#66)
- chore: update benchmark baseline [skip ci]
- deps(deps): bump sysinfo from 0.32.1 to 0.37.2 (#60)
- deps(deps): bump wasmparser from 0.239.0 to 0.240.0 (#64)
- deps(deps): bump governor from 0.6.3 to 0.8.1 (#61)
- chore: update benchmark baseline [skip ci]
- deps(deps): bump mail-parser from 0.9.4 to 0.11.1 (#63)
- deps(deps): bump rumqttc from 0.24.0 to 0.25.0 (#65)
- deps(deps): bump ndarray from 0.16.1 to 0.17.1 (#76)
- chore: update benchmark baseline [skip ci]
- ci(deps): bump azure/setup-helm from 3 to 4 (#72)
- ci(deps): bump actions/upload-artifact from 4 to 5 (#71)
- deps(deps): bump image from 0.24.9 to 0.25.9 (#77)
- deps(deps): bump rustls from 0.21.12 to 0.23.35 (#78)
- chore: update benchmark baseline [skip ci]
- Bump all crates to version 0.3.3
- Format code with rustfmt
- Fix k6 script generation with operation IDs containing dots/hyphens
- chore: update benchmark baseline [skip ci]
- perf: optimize template rendering by avoiding unnecessary operations
- chore: update benchmark baseline [skip ci]
- docs: update benchmark documentation with final optimizations
- perf: fix benchmark regressions and optimize measurements
- chore: update benchmark baseline [skip ci]
- Fix Kafka compilation errors and borrow checker issues
- feat: Implement cross-pillar enhancements - World State Engine, MOD, and Performance Mode
- feat(ai-studio): Add API Critique, System Generator, and Behavioral Simulator
- chore: rework UI/UX to be more AI native
- fix: Address pre-commit security vulnerabilities
- feat: Implement Invisible Mock Server experience (DevX Pillar)
- feat(security): implement email, Slack, and webhook notification services
- Refactor template expansion for Send safety
- chore: Restore path dependencies after 0.3.2 publish
- Fix: Complete SQLx query cache for mockforge-collab 0.3.2
- chore: update mockforge dependencies to version 0.3.1 across multiple crates
- fix: improve dependency conversion for optional dependencies and fix publishing order
- fix: update publish script to handle Phase 1 crate dependencies correctly
- feat: add comprehensive integration tests for 0.3.0 features and update changelog
- feat: Complete pillar enhancement gaps - VS Code extension and docs
- feat: Implement pillar tagging system and documentation enhancements
- feat: Implement MockForge AI Studio - Unified AI Copilot
- feat(cloud): Complete Cloud pillar implementation and fix compilation issues
- [DevX] Add JSON Schema support for config validation and IDE autocompletion
- feat: Implement Contract Fitness Functions, Consumer Impact Analysis, and Multi-Protocol Contracts
- feat: Enhance Reality feature with observability, cross-protocol consistency, and time-aware lifecycles
- fix: use proper vosk API by matching on CompleteResult enum
- fix: resolve all compilation errors
- chore: prepare release 0.3.0
- feat: Implement LLM Studio - Natural Language Workspace Creation (0.3.4)
- feat: Complete Behavioral Cloning v1 implementation and refactor architecture
- feat: Implement Drift Budget & GitOps for API Sync + AI Contract Diff
- feat: implement Scenario Studio Visual Editor with React Flow
- feat: implement AI-Native Interface Deepening features
- feat: Implement Time Travel & Snapshots and Frontend X-Ray Mode
- feat(sdk): Add Contract-Backed Types and Scenario-First SDKs to Vue, Svelte, and Angular
- Format code: Apply rustfmt and whitespace cleanup
- Release v0.2.9: Update version, CHANGELOG, and publish all crates to crates.io
- Add registry server improvements, password reset, metrics, and marketplace enhancements
- security: Upgrade wasmtime to 36.0.3 to fix RUSTSEC-2025-0118
- feat: Fix compilation errors and implement comprehensive E2E test suite
- fix: implement custom routes, template expansion, latency injection, and init improvements
- feat: Smart Personas with array generation and relationship inference
- feat: Complete Java and .NET SDK implementations with builder patterns
- fix: update all test files for new function signatures
- fix: resolve all compilation errors across workspace
- Complete Phase 3 security controls implementation
- Add cloud monetization infrastructure and features
- Implement organization management endpoints
- Fix Axum 0.8 route syntax in state_machine_api.rs
- Fix file server route syntax for Axum 0.8 compatibility
- Release v0.2.8: Publish all crates to crates.io
- chore: bump version to 0.2.8
- feat: Complete Generative Schema Mode and achieve 100% roadmap completion
- Implement Smart Personas feature for consistent cross-endpoint data generation
- Add Reality Continuum feature for blending mock and real data sources
- Implement Voice + LLM Interface with STT backends
- Implement complete Deceptive Deploy feature
- Add GraphQL + REST Playground with workspace filtering
- Implement ForgeConnect SDK with full feature set
- Add enhanced scenario marketplace features
- Configure SQLx and integrate mockforge-collab with mockforge-core
- Fix test compilation errors in reality integration and hot-reload tests
- Implement Reality Slider feature with hot-reload support
- Complete latency recording integration and fix WorkspaceConfig reality_level field
- style: Apply rustfmt formatting to Chaos Lab code
- feat: Add Chaos Lab interactive network condition simulation
- Fix test compilation errors in openapi_generator_tests
- Fix all compilation errors for AI Contract Diff feature
- Add WireMock-inspired features: browser proxy mode, git sync, data sources, template library, managed hosting docs, and user management
- Add comprehensive ecosystem and use cases documentation
- Complete configuration and extensibility implementation
- Add advanced behavior and simulation features
- Fix test and benchmark compilation errors
- Complete Scenario State Machines 2.0 with sub-scenario execution
- Implement VBR Engine enhancements: OpenAPI integration, M2M relationships, seeding, ID generation, snapshots
- Add mock-to-real migration pipeline with per-route toggling
- Add Data Scenarios Marketplace feature
- feat: Implement ForgeConnect - Front-End Integrated Mode for browser-based mock creation
- Add MockForge Cloud Graph visualization with real-time updates and export
- Add data personality profiles system for consistent mock data generation
- Add realistic network conditions and chaos lab with interactive UI controls
- Add temporal simulation with CLI commands and scenario support
- Complete MockAI implementation with query params and session recording
- Add Virtual Backend Reality (VBR) engine
- Add multipart form data support and file generation/serving for API mocks
- fix: update mockforge-plugin-sdk to use workspace version
- fix: enable publishing for mockforge-tunnel and add to publish script
[0.3.20] - 2025-12-31
Fixed
- [Bench] Dynamic placeholder expansion in CRUD flow params file bodies (#79): Fixed
${__VU},${__ITER}, and other dynamic placeholders not being expanded when used in request body content from params files- Previously, placeholders like
"name": "HTTP-WAAP-vsvip-${__VU}-${__ITER}"were sent literally to the API - Now properly converted to k6 template literals for runtime evaluation
- Supports all dynamic placeholders:
${__VU},${__ITER},${__TIMESTAMP},${__UUID},${__RANDOM},${__COUNTER},${__DATE},${__VU_ITER}
- Previously, placeholders like
[0.3.19] - 2025-12-30
Added
- [DevX] API base path support for bench command (#79): New
--base-pathoption to prepend a path prefix to all API endpoints in generated load tests- Automatically extracts base path from OpenAPI spec’s
serversURL (e.g.,https://api.example.com/api/v1→/api/v1) - CLI option takes priority over spec’s base path for explicit control
- Use
--base-path ""to disable base path even if spec defines one - Works with both standard k6 scripts and CRUD flow mode
- Example usage:
# Auto-detect from spec's servers URL mockforge bench --spec api.yaml --target http://localhost:8080 --crud-flow # Explicitly set base path mockforge bench --spec api.yaml --target http://localhost:8080 --base-path /api # Disable base path mockforge bench --spec api.yaml --target http://localhost:8080 --base-path ""
- Automatically extracts base path from OpenAPI spec’s
[0.3.18] - 2025-12-29
Fixed
- [Collab] SQLx offline mode for crates.io installation (#79): Fixed compilation errors when installing
mockforge-collabfrom crates.io- Added
.sqlxquery cache directory with 51 precompiled query metadata files - The
build.rsnow automatically enablesSQLX_OFFLINE=truewhen query cache is present - Users no longer need
DATABASE_URLor to runcargo sqlx prepareto install the crate - Resolves “set DATABASE_URL to use query macros online” compilation errors
- Added
[0.3.17] - 2025-12-28
Added
-
[DevX] WAFBench YAML integration for security testing: New
--wafbench-dirflag to import Microsoft WAFBench CRS (Core Rule Set) attack patterns- Parse WAFBench YAML test files from the WAFBench project
- Support glob patterns for loading specific rule categories (e.g.,
REQUEST-941-*for XSS,REQUEST-942-*for SQLi) - Extract attack payloads from URI parameters, headers, and request bodies
- Automatic CRS rule ID parsing from test metadata (e.g.,
941100for XSS attacks) - Integrate WAFBench payloads with existing security testing framework
- Example usage:
mockforge bench spec.yaml --wafbench-dir ./wafbench/REQUEST-941-* # XSS rules mockforge bench spec.yaml --wafbench-dir ./wafbench/**/*.yaml # All rules
-
[DevX] Per-URI control mode for data-driven testing (#79): New
--per-uri-controlflag for CSV/JSON data files that allows each row to specify HTTP method, URI, body, query params, headers, attack type, and expected status code- Enables fine-grained control over test requests directly from data files
- Supports security testing per-URI with
attack_typecolumn - Automatic status validation with
expected_statuscolumn - Example CSV format:
method,uri,body,query_params,headers,attack_type,expected_status GET,/virtualservice,,include_name=true,,,200 POST,/virtualservice,"{""name"":""test""}",,,sqli,201
-
[Protocol] AMQP TLS support: Full TLS/SSL support for AMQP broker with configurable certificates
-
[Protocol] MQTT protocol improvements: Enhanced MQTT server with TLS, session management, and metrics
-
[Protocol] gRPC dynamic service improvements: Better dynamic proto loading and error handling
-
[Registry] Security enhancements: CSRF protection, request ID middleware, trusted proxy support, token revocation
-
[UI] Frontend improvements: Type safety fixes, mobile layout improvements, search/filter functionality
Changed
- Comprehensive dependency updates across workspace crates
Fixed
- [DevX] CRUD flow params file integration (#79): Fixed
--params-filenot being applied in CRUD flow mode- Body configurations from params file are now correctly applied to POST/PUT/PATCH operations in
--crud-flowmode - Fixed body serialization issue that caused “ReferenceError: object is not defined” error in generated k6 scripts
- Body is now properly serialized as a JSON string for the Handlebars template
- Body configurations from params file are now correctly applied to POST/PUT/PATCH operations in
- [Core] Race conditions and flaky tests: Resolved timing issues across test suite
- [Core] Panic-prone unwrap calls: Replaced with safe error handling throughout codebase
[0.3.16] - 2025-12-27
Added
- Version bump with dependency updates
Fixed
- [Test] Flaky test fixes: Resolved race conditions and timing issues in integration tests
- [Core] Safe error handling: Replaced panic-prone
.unwrap()calls with proper error handling
[0.3.15] - 2025-12-26
Added
- [DevX] Multi-spec support for bench command: The
mockforge benchcommand now supports loading and merging multiple OpenAPI specifications- Multiple
--specflags:mockforge bench --spec pools.yaml --spec vs.yaml --target https://api.com - Directory discovery with
--spec-dir:mockforge bench --spec-dir ./specs/ --target https://api.com - Conflict resolution strategies with
--merge-conflicts:error(default),first,last - Spec mode selection with
--spec-mode:merge(default) combines all specs,sequentialruns specs in dependency order - Sequential execution mode with per-spec output directories and results
- Leverages existing multi-spec infrastructure from mockforge-core
- Multiple
- [DevX] Cross-spec dependency detection: New
spec_dependenciesmodule for handling dependencies between specs- Automatic detection of dependencies from field naming patterns (
pool_ref,pool_id,poolId, etc.) - Schema registry for cross-referencing schemas across multiple specs
- Topological sorting for correct execution order
- Manual dependency configuration via
--dependency-config(YAML/JSON) - Support for value extraction and injection between spec groups
- Automatic detection of dependencies from field naming patterns (
Changed
BenchCommand.specfield changed fromPathBuftoVec<PathBuf>to support multiple specsSpecParsernow includesfrom_spec()method for pre-loaded OpenAPI specs- Added
dependency_configfield toBenchCommandfor cross-spec value passing configuration
Fixed
- Nothing yet.
[0.3.14] - 2025-12-26
Added
- Version bump to 0.3.14
Changed
- Nothing yet.
Fixed
- Nothing yet.
[0.3.13] - 2025-12-24
Fixed
- [DevX] k6 CRUD flow metric name sanitization (#79 follow-up): Fixed invalid k6 metric names in CRUD flow scripts when flow names contain dots or special characters
- CRUD flow names are now sanitized for use as k6 metric names (e.g.,
plans.list→plans_list) - Original flow names preserved in comments and group names for readability
- Made
sanitize_js_identifierfunction public for reuse across k6 generators - Added script validation to CRUD flow generation for defense in depth
- CRUD flow names are now sanitized for use as k6 metric names (e.g.,
[0.3.12] - 2025-12-23
Changed
- [DevX] Dependency updates: Version alignment and dependency updates across all workspace crates
[0.3.11] - 2025-12-19
Added
-
[DevX] Custom benchmark parameters: Added
--params-fileoption tomockforge benchcommand for loading custom parameter values from a fileWhy it matters: Allows users to define reusable parameter configurations for benchmark runs, making it easier to test different scenarios without modifying command-line arguments each time.
[0.3.10] - 2025-12-18
Fixed
- [DevX] k6 benchmark script generation fixes: Resolved multiple issues with generated k6 scripts
- Fixed
--insecureflag handling by movinginsecureSkipTLSVerifyto global k6 options - Fixed
textSummaryimport and usage in generated scripts - Fixed query parameter encoding in benchmark requests
- Fixed
[0.3.9] - 2025-12-17
Added
-
[Reality] Comprehensive Mock Server Implementation: Full implementation across all protocol crates
- mockforge-amqp: Complete AMQP 0-9-1 broker with exchanges, queues, bindings, messages, protocol handling, fixtures, and spec registry
- mockforge-kafka: Full Kafka broker with consumer groups, partitions, topics, metrics, and protocol handling
- mockforge-mqtt: Complete MQTT broker with QoS levels, topic subscriptions, and retained messages
- mockforge-ftp: Virtual filesystem, spec registry, and fixture support
- mockforge-smtp: Email server with fixtures and spec registry
- mockforge-tcp: TCP server with fixtures and protocol support
- mockforge-grpc: Dynamic proto parser, service generator, reflection, and metrics
- mockforge-graphql: Full handler implementations
-
[DevX] Enhanced CLI Commands: New commands for all protocols and features
- AMQP, Kafka, MQTT, FTP, SMTP protocol commands
- Blueprint, cloud, deploy, dev-setup, governance commands
- Logs, progress, recorder, scenario, snapshot commands
- Time manipulation, VBR, voice, wizard, and workspace commands
- AI-powered mock generation commands
-
[Reality] Virtual Backend Repository (VBR): Complete data management system
- API generator, entity management, constraints, and validation
- Database integration with migrations and schema management
- Session handling, snapshots, and mutation rules
- ID generation strategies and scheduling
-
[Reality] World State Engine: Coherent world simulation
- State engine with model and query support
- Entity relationships and lifecycle management
-
[AI] Enhanced AI Capabilities: AI-powered mock generation
- RAG-based AI response generator
- AI event generator for WebSocket scenarios
- Behavioral cloning with scenario types
-
[Cloud] Collaboration Features: Team collaboration support
- Backup, merge, and promotion workflows
- Multi-environment configuration
- Client SDK improvements
-
[DevX] Observability & Analytics: Enhanced monitoring
- Pillar usage tracking and analytics queries
- Metrics middleware and coverage tracking
- Latency metrics and performance monitoring
-
[Contracts] Chaos Engineering: Resilience testing capabilities
- Failure designer and incident replay
- Chaos API with configurable fault injection
- Route-level chaos with latency distributions
-
[DevX] Plugin System Enhancements: Extended plugin capabilities
- Backend generator and datasource support
- Runtime adapter improvements
- SDK builders and testing utilities
-
[Cloud] Registry Server: Complete registry implementation
- Authentication, authorization, and RBAC
- Redis caching, email notifications
- Organization and subscription models
- API token management and audit logging
-
[DevX] UI Server: Dashboard and admin features
- Admin handlers for workspace management
- Chain visualization and coverage metrics
- Failure analysis and promotion workflows
- Graph visualization and health monitoring
[0.3.8] - 2025-01-27
Fixed
-
[DevX] Compilation errors resolved: Fixed all compilation errors across the workspace
- Updated
axum-serverfrom 0.6 to 0.8 withtls-rustls-no-providerfeature - Updated
rustlsfrom 0.21 to 0.23,rustls-pemfilefrom 1.0 to 2.0,tokio-rustlsfrom 0.24 to 0.26 - Adapted TLS code to rustls 0.23 API (CertificateDer, PrivateKeyDer, WebPkiClientVerifier)
- Fixed multi_spec module: properly exported and resolved compilation errors
- Fixed handle_serve function calls: added missing parameters and fixed type mismatches
- Fixed borrow checker issues in multi_spec merging logic
- Added missing documentation for enum variants and struct fields
- Fixed various type mismatches and iteration patterns
- Updated
-
[DevX] Cargo publish readiness: Fixed all dependency version requirements for crates.io publishing
- Added version requirements to all path dependencies in mockforge-cli, mockforge-chaos, mockforge-http, mockforge-route-chaos, mockforge-vbr
- Set
publish = falsefor desktop-app and tests packages (not meant for crates.io) - All crates now pass
cargo publish --dry-runvalidation
[0.3.6] - 2025-11-25
Fixed
-
[DevX] k6 script generation with operation IDs containing dots/hyphens (#79)
- Fixed “Unexpected token .” error when OpenAPI operation IDs contain dots (e.g.,
plans.create) or hyphens (e.g.,plans.update-pricing-schemes) - Changed
is_alphanumeric()tois_ascii_alphanumeric()in JavaScript identifier sanitization to ensure ASCII-only identifiers - All operations are now properly included in generated k6 scripts with valid JavaScript identifiers
- Added comprehensive tests including integration test with full billing subscriptions spec
- Fixed “Unexpected token .” error when OpenAPI operation IDs contain dots (e.g.,
-
[DevX] UI icon embedding for published crates
- Fixed build failures when installing
mockforge-clifrom crates.io due to missing icon files - Updated
build.rsto read icon files at build time and embed them as byte array literals - Replaced
include_bytes!withCARGO_MANIFEST_DIRapproach that failed in published crates - Icons are now properly embedded and work both in development and when installing from crates.io
- Fixed build failures when installing
[0.3.0] - 2025-11-17
Added
-
[DevX] Pillars & Tagged Changelog: Complete pillar system implementation with documentation and tooling
- Defined five foundational pillars: [Reality], [Contracts], [DevX], [Cloud], [AI]
- Added comprehensive PILLARS.md documentation with feature mappings
- Implemented CI validation for pillar tags in changelog entries
- Added pillar tagging instructions to release tooling
- Updated README and getting-started guide with pillars section
Why it matters: Clear product story spine that makes it obvious what each release invests in. Pillar tags help users understand product direction and find features relevant to their needs.
-
[Reality] Smart Personas & Reality Continuum v2: Complete persona graph and lifecycle system
- Persona graphs with relationship linking across entities
- Lifecycle states (NewSignup, Active, PowerUser, ChurnRisk, Churned, etc.)
- Reality Continuum integration with field-level and entity-level mixing
- Fidelity score calculation and API endpoint
- Comprehensive PERSONAS.md documentation
Why it matters: Upgrade from “random-but-consistent fake data” to “coherent world simulation.” Personas maintain relationships across endpoints, and fidelity scores quantify how real your mock environment is.
-
[Contracts] Drift Budget & GitOps for API Sync: Complete drift management system
- Hierarchical drift budget configuration (global, workspace, service, endpoint)
- Breaking change detection and classification
- Incident management with webhook integration
- GitOps PR generation for contract updates
- Comprehensive DRIFT_BUDGETS.md documentation
Why it matters: Make MockForge the “drift nerve center” for contracts. Define acceptable drift, get alerts when budgets are exceeded, and automatically generate PRs to update contracts and fixtures.
-
[Reality] Behavioral Cloning v1: Multi-step flow recording and replay
- Flow recording with request/response capture and timing
- Flow viewer with timeline visualization
- Scenario replay engine with strict/flex modes
- Scenario storage and export/import (YAML/JSON)
- Comprehensive BEHAVIORAL_CLONING.md documentation
Why it matters: Move from endpoint-level mocks to journey-level simulations. Record realistic flows from real systems and replay them as named scenarios for comprehensive testing.
-
[AI][DevX] LLM/Voice Interface for Workspace Creation: Natural language to complete workspace
- Natural language workspace creation from descriptions
- Automatic persona and relationship generation
- Behavioral scenario generation (happy path, failure, slow path)
- Reality continuum and drift budget configuration from NL
- Voice and text input support
- Comprehensive LLM Studio documentation
Why it matters: The golden path: “Describe the system in natural language → MockForge builds a realistic mock backend with personas, behaviors, and reality level config.” No manual configuration required.
-
[DevX] Comprehensive Integration Test Coverage: Complete test suite for all 0.3.0 features
- Smart Personas v2 integration tests (15 tests covering persona graphs, lifecycle states, fidelity scores)
- Drift Budget integration tests (14 tests covering budget hierarchy, breaking change detection, incident management)
- Drift GitOps integration tests (16 tests covering PR generation, OpenAPI/fixture updates, GitOps configuration)
- Behavioral Cloning integration tests (15 tests covering flow recording, scenario replay, strict/flex modes)
- Voice/LLM Workspace Creation integration tests (16 tests covering command parsing, workspace building, NL to workspace flow)
- All tests passing with 100% success rate (76 total integration tests)
Why it matters: Production-ready features require production-ready tests. Comprehensive integration test coverage ensures reliability, prevents regressions, and provides confidence for users adopting these features.
Changed
- Changelog entries now require pillar tags for all major features
- Release process includes automated pillar tag validation
- Documentation structure updated to highlight pillars
Fixed
- Nothing yet.
Security
- Nothing yet.
[0.2.9] - 2025-11-14
Added
-
[Cloud] Registry server improvements with password reset functionality
Why it matters: Enable seamless team collaboration with secure registry access—teams can share and discover mock scenarios without friction, and password reset keeps workflows moving when credentials are lost.
-
[Cloud] Enhanced metrics and marketplace features
-
[DevX] Comprehensive E2E test suite
-
[DevX] Custom routes implementation
-
[Reality] Template expansion improvements
-
[Reality] Latency injection enhancements
-
[Reality] Smart Personas with array generation and relationship inference
Why it matters: Generate realistic, interconnected mock data automatically—arrays that make sense, relationships that stay consistent across endpoints, and personas that feel like real users without manual configuration.
-
[DevX] Complete Java and .NET SDK implementations with builder patterns
Why it matters: Bring MockForge to enterprise teams using Java and .NET—no more language barriers, no more custom integration work. Your entire stack can use the same mock infrastructure.
-
[Cloud] Cloud monetization infrastructure and features
Why it matters: Enable sustainable platform growth with flexible pricing models—teams can scale from free tier to enterprise without friction, and the platform can grow while serving developers.
-
[Cloud] Organization management endpoints
Why it matters: Scale from solo developer to enterprise team—manage users, permissions, and resources at the org level, not just individual accounts. Real teams need real organization tools.
-
[Cloud] Security controls implementation (Phase 3)
Why it matters: Protect production deployments with enterprise-grade security—fine-grained access controls, audit trails, and compliance features that let you trust MockForge with sensitive data and critical workflows.
Changed
- [DevX] Upgraded wasmtime to 36.0.3 to fix RUSTSEC-2025-0118
- [DevX] Fixed Axum 0.8 route syntax compatibility across multiple modules
- [DevX] Updated all test files for new function signatures
Fixed
- [DevX] Fixed compilation errors across workspace
- [DevX] Fixed Axum 0.8 route syntax in state_machine_api.rs
- [DevX] Fixed file server route syntax for Axum 0.8 compatibility
- [DevX] Resolved all compilation errors for comprehensive test coverage
Security
- [DevX] Upgraded wasmtime to 36.0.3 to address RUSTSEC-2025-0118
- [Cloud] Completed Phase 3 security controls implementation
[0.2.8] - 2025-11-10
Added
-
[Reality] Generative Schema Mode: Complete implementation of generative schema mode for dynamic mock data generation
Why it matters: Spin up a believable API even when the backend doesn’t exist yet—no sample DB or seed data required.
-
[Reality] Smart Personas: Feature for consistent cross-endpoint data generation using persona-based templates
-
[Reality] Reality Continuum: Feature for blending mock and real data sources with configurable reality levels
Why it matters: Turn the dial between deterministic mock and noisy production-like chaos without changing your client code.
-
[Reality] Reality Slider: Hot-reload support for reality level adjustments
Why it matters: Adjust reality levels on the fly during development and testing without restarting the server.
-
[Reality] Chaos Lab: Interactive network condition simulation tool
Why it matters: Test how your application handles real-world network conditions like latency spikes, packet loss, and connection failures.
-
[Contracts] AI Contract Diff: Feature for comparing and diffing API contracts
Why it matters: Automatically detect and visualize API contract changes to catch breaking changes before they reach production.
-
[DevX] Voice + LLM Interface: Voice interface implementation with Speech-to-Text (STT) backend support
-
[Reality] Deceptive Deploy: Complete deceptive deploy feature for advanced testing scenarios
-
[DevX] GraphQL + REST Playground: Interactive playground with workspace filtering capabilities
-
[DevX] ForgeConnect SDK: Complete SDK implementation with full feature set
-
[Cloud] Enhanced Scenario Marketplace: Improved scenario marketplace with additional features
-
[DevX] WireMock-Inspired Features: Browser proxy mode, git sync, data sources, template library, managed hosting documentation, and user management
-
[DevX] Ecosystem Documentation: Comprehensive ecosystem and use cases documentation
-
[DevX] Configuration Extensibility: Complete configuration and extensibility implementation
-
[Reality] Advanced Behavior Simulation: Enhanced behavior and simulation features
Changed
- [DevX] SQLx Integration: Configured SQLx and integrated mockforge-collab with mockforge-core
- [Reality] Latency Recording: Completed latency recording integration with WorkspaceConfig reality_level field support
Fixed
- [DevX] Fixed test compilation errors in reality integration and hot-reload tests
- [DevX] Fixed test compilation errors in openapi_generator_tests
- [Contracts][DevX] Fixed all compilation errors for AI Contract Diff feature
- [DevX] Applied rustfmt formatting to Chaos Lab code
Security
- Nothing yet.
[0.2.7] - 2025-11-05
Added
-
[Contracts] Automatic API Sync & Change Detection: Implemented periodic polling and automatic sync for detecting upstream API changes
Why it matters: Keep your mocks in sync with real APIs automatically—catch breaking changes before they break your tests.
- Periodic sync service with configurable intervals (default: 1 hour)
- Automatic change detection using deep response comparison (status, headers, body)
- Optional automatic fixture updates when changes detected
- Manual sync trigger via API (
POST /api/recorder/sync/now) - Sync status tracking and change history
- Configurable sync settings: upstream URL, interval, headers, timeout, max requests
- Support for GET-only or all-methods sync
- Detailed change reports with before/after comparisons
- Database update method for refreshing recorded responses
- API endpoints:
/api/recorder/sync/status,/api/recorder/sync/config,/api/recorder/sync/changes
-
[Reality] TCP Protocol Support: Added raw TCP server mocking support via new
mockforge-tcpcrateWhy it matters: Mock any protocol that runs over TCP—not just HTTP. Perfect for testing database clients, custom protocols, and legacy systems.
- Raw TCP connection handling with fixture-based matching
- Echo mode for testing TCP clients
- TLS/SSL support for encrypted connections
- Delimiter-based message framing (optional)
- Configurable buffer sizes and connection limits
- CLI flag
--tcp-portfor custom TCP server port - Configuration via
config.tcpin YAML/JSON config files
-
[Reality] Response Selection Modes: Added support for sequential (round-robin) and random response selection when multiple examples are available
- Sequential mode: Cycles through available examples in order (round-robin)
- Random mode: Randomly selects from available examples
- Weighted random mode: Random selection with custom weights per example
- Configuration via
x-mockforge-response-selectionOpenAPI extension - Environment variable support:
MOCKFORGE_RESPONSE_SELECTION_MODE(global) andMOCKFORGE_RESPONSE_SELECTION_<OPERATION_ID>(per-operation) - State tracking for sequential mode ensures round-robin behavior across requests
-
[Reality] Webhook HTTP Execution: Implemented actual HTTP request execution in chaos orchestration hooks
HookAction::HttpRequestnow executes real outbound HTTP requests (previously only logged)- Supports GET, POST, PUT, DELETE, PATCH methods
- Configurable request body and headers
- Error handling and logging for webhook failures
- Fire-and-forget execution (failures don’t block orchestration)
-
[DevX] CRUD & Webhook Documentation: Added comprehensive documentation guides
docs/CRUD_SIMULATION.md: Complete guide for simulating CRUD operations with stateful data storedocs/WEBHOOKS_CALLBACKS.md: Full documentation of webhook capabilities via hooks, chains, and scripts- Examples demonstrating realistic workflows and integrations
Changed
- Nothing yet.
Deprecated
- Nothing yet.
Removed
- Nothing yet.
Fixed
- Nothing yet.
Security
- Nothing yet.
[0.2.6] - 2025-11-04
Added
-
[DevX] TLS/HTTPS and mTLS Support: Added TLS/HTTPS and mutual TLS (mTLS) support for HTTP server
- Configurable TLS certificate and key paths
- Client certificate authentication support
- Secure connection handling for production deployments
-
[DevX] Built-in Tunneling Service: Added built-in tunneling service for exposing local servers via public URLs
- Automatic tunnel creation for local development
- Public URL generation for testing and demos
- Integration with popular tunneling services
-
[DevX] SDK Implementation: Completed Phase 1 & 2 of SDK implementation
- Comprehensive documentation and examples
- Production-ready client generators
Changed
-
[DevX] Version Bumps: Updated all workspace crates from 0.2.5 to 0.2.6
- Updated all dependency versions across the workspace
- Fixed version mismatches in mockforge-ui and mockforge-plugin-loader
-
[DevX] Publishing Improvements: Enhanced crate publishing process
- Added mockforge-tcp and mockforge-test to publish script
- Enabled publishing for mockforge-test crate
- Fixed mockforge-tcp to remove README requirement
Fixed
-
[DevX] Documentation: Fixed missing module-level documentation in test files
- Added comprehensive module documentation to all test modules
- Improved code documentation consistency
-
[DevX] Axum Compatibility: Fixed Axum 0.8 compatibility issues in proxy server module
- Updated proxy server to work with latest Axum version
- Resolved breaking changes from Axum upgrade
-
[Reality] MQTT Error Types: Fixed MQTT publish handlers error types to be Send + Sync
- Updated error types for proper async/await compatibility
- Ensured thread-safety in MQTT handlers
[0.2.5] - 2025-01-27
Added
-
[DevX] OAuth2 Flow Support: Complete OAuth2 implementation with all standard flows
- Authorization Code flow with PKCE (RFC 7636 compliant, SHA256 hash)
- Client Credentials flow for server-side applications
- Password flow for trusted clients
- Implicit flow support
- Automatic token refresh and expiration management
- State parameter for CSRF protection
- PKCE code verifier/challenge generation helpers
- Token storage with expiration tracking (localStorage)
-
[DevX] Enterprise Error Handling: Structured error handling for generated clients
ApiErrorclass with status codes, statusText, and error bodyRequiredErrorclass for missing required fields- Helper methods:
isClientError(),isServerError(),getErrorDetails(),getVerboseMessage() - Optional verbose error messages with detailed validation information
-
[Contracts] Request/Response Validation: Built-in validation support
- Required field validation before sending requests
- Basic response structure validation (type checking, object validation)
- Configurable via
validateRequestsflag - Detailed validation error messages
-
[DevX] Request/Response Interceptors: Custom request/response/error transformation
- Request interceptor: Modify requests before sending
- Response interceptor: Transform responses after receiving
- Error interceptor: Global error handling
- Support for async interceptors
-
[DevX] Enhanced Authentication: Multiple authentication methods
- Bearer token (static or dynamic function)
- API key authentication (static or dynamic)
- Basic authentication (username/password)
- OAuth2 (all flows, takes priority over other methods)
-
[DevX] PKCE Helper Functions: Exported utilities for PKCE implementation
generatePKCECodeVerifier(): Generate cryptographically random code verifiergeneratePKCECodeChallenge(): Generate SHA256 code challenge from verifier
-
[DevX] Security Best Practices: Comprehensive security warnings and guidance
- Client secret warnings for browser-based applications
- XSS vulnerability warnings for localStorage token storage
- CSRF protection via state parameter validation
- Token expiration checking
- Security documentation in generated README
-
[DevX] Request Timeout Handling: Configurable request timeouts
- Default 30-second timeout (configurable)
- AbortController-based timeout implementation
- Proper timeout error handling
-
[DevX] React Query Integration Documentation: Comprehensive examples for @tanstack/react-query integration
Changed
-
[DevX] React Client Generator: Major enhancements to generated React client code
- Replaced placeholder PKCE implementation with full SHA256-based solution
- Implemented proper response validation (previously placeholder)
- Enhanced README with comprehensive feature documentation
- Improved error messages and validation details
- Better security documentation and best practices
-
[DevX] Operation ID Sanitization: Improved identifier generation
- Enhanced
sanitize_identifierfunction to handle complex operation IDs - Better handling of parentheses, slashes, hyphens in operation IDs
- Proper camelCase conversion with word boundary detection
- Enhanced
Fixed
-
[DevX] TypeScript Empty Object Types: Fixed formatting issue where empty object schemas generated invalid TypeScript
- Empty objects now correctly generate as
[key: string]: any;instead of malformedRecord<string, any>}
- Empty objects now correctly generate as
-
[DevX] DELETE Operations with Query Params: Fixed missing query parameter support in DELETE operations
-
[DevX] Duplicate Operation IDs: Fixed duplicate operation ID handling by appending numeric suffixes
-
[DevX] PKCE Code Challenge: Fixed PKCE implementation to use proper SHA256 hash instead of plain encoding
-
[Contracts][DevX] Response Validation: Replaced placeholder with actual implementation (type checking, structure validation)
Security
- [DevX] Added comprehensive security warnings for OAuth2 client secrets in browser code
- [DevX] Added XSS vulnerability warnings for localStorage token storage
- [DevX] Implemented CSRF protection via state parameter validation
- [DevX] Added token expiration checking to prevent use of expired tokens
- [DevX] Documented security best practices in generated client README
[0.2.4] - 2025-01-27
Fixed
- [DevX] Fix request body parameter generation in React/Vue/Svelte client generators - request bodies now correctly generate
dataparameter andbody: JSON.stringify(data)in API client methods - [DevX] Fix required vs optional field handling in generated TypeScript interfaces - required fields no longer incorrectly marked with optional marker (
?) - [DevX] Fix OpenAPI serde deserialization by adding
#[serde(rename)]attributes foroperationIdandrequestBodyfields - [DevX] Apply required fields processing consistently across all client generators (React, Vue, Svelte)
Added
- [DevX] Comprehensive test coverage for request body parameter scenarios (POST, PUT, PATCH, DELETE)
- [DevX] Test cases for
$refschemas in request bodies - [DevX] Test cases for YAML spec support verification
[0.2.3] - 2025-01-27
Fixed
- [DevX] Fix OpenAPI example extraction to prioritize explicit examples from schema and properties
- [DevX] Fix request body parameter generation in React client generator for POST, PUT, PATCH, DELETE methods
- [DevX] Fix Handlebars template logic for request body type generation in client code
- [DevX] Fix useCallback dependency array formatting in React hooks template
- [DevX] Add comprehensive test coverage for request body parameter scenarios
[0.2.0] - 2025-10-29
Added
- [DevX] Output control features for MockForge generator with comprehensive configuration options
- [DevX] Unified spec parser with enhanced validation and error reporting
- [DevX] Multi-framework client generation with Angular and Svelte support
- [Reality] Enhanced mock data generation with OpenAPI support
- [DevX] Configuration file support for mock generation
- [DevX] Browser mobile proxy mode implementation
- [DevX] Comprehensive documentation and example workflows
Changed
- [DevX] Enhanced CLI with progress indicators, error handling, and code quality improvements
- [DevX] Comprehensive plugin architecture documentation
Fixed
- [DevX] Remove tests that access private fields in mock data tests
- [DevX] Fix compilation issues in mockforge-collab and mockforge-ui
- [DevX] Update mockforge-plugin-core version to 0.1.6 in plugin-sdk
- [DevX] Enable SQLx offline mode for mockforge-collab publishing
- [DevX] Add description field to mockforge-analytics
- [DevX] Add version requirements to all mockforge path dependencies
- [DevX] Fix publish order dependencies (mockforge-chaos before mockforge-reporting)
- [DevX] Update Cargo.lock and format client generator tests
[0.1.3] - 2025-10-22
Changes
- [DevX] docs: prepare release 0.1.3
- [DevX] docs: update CHANGELOG for 0.1.3 release
- [DevX] docs: add roadmap completion summary
- [DevX] feat: add Kubernetes-style health endpoint aliases and dashboard shortcut
- [DevX] feat: add unified config & profiles with multi-format support
- [Reality] feat: add capture scrubbing and deterministic replay
- [DevX] feat: add native GraphQL operation handlers with advanced features
- [Reality] feat: add programmable WebSocket handlers
- [Reality] feat: add HTTP scenario switching for OpenAPI response examples
- [DevX] feat: add mockforge-test crate and integration testing examples
- [DevX] build: enable publishing for mockforge-ui and mockforge-cli
- [DevX] build: extend publish script for internal crates
- [DevX] build: parameterize publish script with workspace version
[0.1.2] - 2025-10-17
Changes
- [DevX] build: make version update tolerant
- [DevX] build: manage version references via wrapper
- [DevX] build: mark example crates as non-publishable
- [DevX] build: drop publish-order for cargo-release 0.25
- [DevX] build: centralize release metadata in release.toml
- [DevX] build: remove per-crate release metadata
- [DevX] build: fix release metadata field name
- [DevX] build: move workspace release metadata into Cargo.toml
- [DevX] build: require execute flag for release wrapper
- [DevX] build: automate changelog generation during release
- [DevX] build: add release wrapper with changelog guard
- [DevX] build: align release tooling with cargo-release 0.25
[0.1.1] - 2025-10-17
Added
- [Contracts] OpenAPI request validation (path/query/header/cookie/body) with deep $ref resolution and composite schemas (oneOf/anyOf/allOf).
- [Contracts] Validation modes:
disabled,warn,enforce, with aggregate error reporting and detailed error objects. - [DevX] Runtime Admin UI panel to view/toggle validation mode and per-route overrides; Admin API endpoint
/__mockforge/validation. - [DevX] CLI flags and config options to control validation (including
skip_admin_validationand per-routevalidation_overrides). - [DevX] New e2e tests for 2xx/422 request validation and response example expansion across HTTP routes.
- [DevX] Templating reference docs and examples; WS templating tests and demo update.
- [Reality] Initial release of MockForge - Multi-protocol mocking framework
- [Reality] HTTP API mocking with OpenAPI support
- [Reality] gRPC service mocking with Protocol Buffers
- [Reality] WebSocket connection mocking with replay functionality
- [DevX] CLI tool for easy local development
- [DevX] Admin UI for managing mock servers
- [DevX] Comprehensive documentation with mdBook
- [DevX] GitHub Actions CI/CD pipeline
- [DevX] Security audit integration
- [DevX] Pre-commit hooks for code quality
Changed
- [Contracts] HTTP handlers now perform request validation before routing; invalid requests return 400 with structured details (when
enforce). - [Contracts] Bump
jsonschemato 0.33 and adapt validator API; enable draft selection and format checks internally. - [Contracts] Improve route registry and OpenAPI parameter parsing, including styles/explode and array coercion for query/header/cookie parameters.
Deprecated
- N/A
Removed
- N/A
Fixed
- [DevX] Resolve admin mount prefix from config and exclude admin routes from validation when configured.
- [Contracts] Various small correctness fixes in OpenAPI schema mapping and parameter handling; clearer error messages.
Security
- N/A
Release Process
This project uses cargo-release for automated releases.
Creating a Release
-
Patch Release (bug fixes):
make release-patch -
Minor Release (new features):
make release-minor -
Major Release (breaking changes):
make release-major
Manual Release Process
If you need to do a manual release:
- Update version in
Cargo.tomlfiles - Update
CHANGELOG.mdwith release notes - Commit changes:
git commit -m "chore: release vX.Y.Z" - Tag:
git tag vX.Y.Z - Push:
git push && git push --tags - Publish to crates.io:
cargo publish
Pre-release Checklist
-
All tests pass (
make test) -
Code formatted (
make fmt) -
Lints pass (
make clippy) -
Security audit passes (
make audit) - Documentation updated
- Changelog updated
-
Version bumped in all
Cargo.tomlfiles - Breaking changes documented (if any)
- CI passes on all branches