reject non-string model with 400 before resolver (#6407, 6/6). Reconciled with #6437 early schema validation on chat.ts. Integrated into release/v3.8.46.
Dynamic MITM manager import on the agent-bridge route + Turbopack static-analyzer anchor in the skills generator (#6329). Integrated into release/v3.8.46.
* fix(docker): make @/mitm/manager Turbopack stub opt-in (OMNIROUTE_MITM_STUB=1) so npm/Electron/VPS builds bundle the real manager (#6344)
v3.8.45 flipped the production bundler default to Turbopack. next.config.mjs
aliased @/mitm/manager to the Docker-only degraded stub unconditionally, which
was harmless while Docker was the only Turbopack consumer but shipped the stub
to every npm/Electron/VPS artifact once Turbopack became the default — breaking
Agent Bridge start with 'MITM manager stub reached at runtime'. The alias is now
gated on OMNIROUTE_MITM_STUB=1 (set only by the Dockerfile) via the shared
scripts/build/mitm-stub-flag.mjs helper.
Regression guard: tests/unit/mitm-stub-alias-6344.test.mjs (4).
* test(next-config): align mitm-manager alias assertion with opt-in stub (#6344)
The existing test asserted the alias was always present; #6344 makes it opt-in
(OMNIROUTE_MITM_STUB=1). Default build now asserts no alias, plus a new env-matrix
test covering both the default (no stub) and Docker (stub) cases.
* docs(changelog): restore #6359 bullet eaten by merge auto-resolve
Clients like the Codex app emit parameters:{type:null,...} for some tools;
OpenAI-compatible upstreams reject with '400 Invalid schema for function ...:
schema must be a JSON Schema of type object, got type null'. toolSchemaSanitizer
already dropped the null; it now re-adds the mandatory root object type (plus
empty properties / open additionalProperties when absent). Combinator roots
(anyOf/oneOf/allOf) and explicit root types are preserved.
Regression guard: 5 new cases in tests/unit/tool-schema-sanitizer.test.mjs.
Two defects found live in the v3.8.45 Phase 5 run (2026-07-06):
- execFileSync default 1 MiB maxBuffer crashed on git show origin/main:CHANGELOG.md
- the i18n resync only synced the [NEXT] (TBD) section, leaving the just-shipped
finalized section as '— TBD' in all 42 mirrors; now also syncs [prevVersion]
bounded by the heading below it (new pure helper versionAfter, unit-tested)
* test(ci): quarantine concurrency-sensitive flakes into a serial pass (tests/unit/serial/)
glm-coding-plan-monthly-3580, quota-division-blocks and provider-health-autopilot
fail under --test-concurrency>1 CPU contention but pass isolated (v3.8.45 release
benchmark: this class cost ~28min re-runs per CI round). They now live in
tests/unit/serial/ and run in a dedicated --test-concurrency=1 step appended to
every runner (test:unit, :ci, :ci:shard, :fast, :shard:1/2, coverage:runner —
sharded variants shard the serial pass too, so concurrent shard jobs never
self-collide). Discovery + TIA gates know the new glob; stryker.conf.json path
updated. Guard: tests/unit/test-serial-quarantine.test.ts (4 tests).
* test(ci): quarantine combo-health-autopilot too (async logger writes after test end under load)
Fresh evidence from this PR's own CI: all 3 subtests pass but an async log write
lands after the test ends (ENOENT app.log -> uncaughtException) — same
concurrency-flake family. Moved to tests/unit/serial/, imports adjusted,
stryker path updated, guard test now asserts 4 quarantined files.
The pre-flight ran unit (~25-35min) + vitest (~3-8min) + integration (~3-10min)
+ pack-artifact (~15min) SEQUENTIALLY via execFileSync — ~1h wall in the
v3.8.45 release (the dominant cost of Phase 0). They are independent processes
with per-process DATA_DIR isolation, so they now run in a single Promise.all
over an async runAsync() twin of run(): total drops to ~the slowest single
suite (~30min). Each still saves its per-gate log (_artifacts/release-green/)
and reports HARD independently. --quick still skips them; --with-build folds
pack-artifact into the same wave (or runs it alone under --quick).
Guard: tests/unit/validate-release-green.test.ts (13/13).
Incident 2026-07-05 on the self-hosted release runner (VM 113): the
integration test 'POST /cert: installs trust when cert exists' exercised the
REAL install path, wrote a 105-byte fake PEM (FakeMITMCertForTestingOnly)
into /usr/local/share/ca-certificates and update-ca-certificates baked the
invalid entry into ca-certificates.crt — breaking ALL system TLS on the VM
(curl error 77, apt cert failures, and the intermittent gzip-corrupted
next-build artifacts that failed 6/9 e2e shards in run 28754447912). Hosted
runners are ephemeral, so the same mutation went unnoticed for months.
- installCert/uninstallCert: skip the OS dispatch under
OMNIROUTE_SKIP_SYSTEM_TRUST=1 — AFTER the input checks, so the #4546
environment-skip contract (missing file throws -> structured skip) and the
already-installed/not-installed early returns are preserved.
- installTproxyCa/uninstallTproxyCa: same guard, only when no run dep is
injected (DI'd tests keep exercising the full command sequence with mocks).
- tests/_setup/isolateDataDir.ts sets the env for every node:test process;
ci.yml/quality.yml/nightly-release-green.yml set it workflow-wide (e2e runs
the real app outside the test setup).
TDD: tests/unit/system-trust-test-guard.test.ts (guard exported to every test
process; guarded install resolves on a real file without touching the OS;
missing-file contract preserved). 82/82 across the affected cert/tproxy/
agent-bridge suites.
Schemathesis's newer unsupported-methods check (unpinned tool drift) sends
QUERY /api/keys/{id}/devices and demands 405 Method Not Allowed; the path had
no HIGH_RISK_METHOD_RULES entry, so the auth layer answered 401 first. Add
the devices rule (GET-only) so undocumented methods get a clean method-first
405 — same pattern as the v3.8.44 TRACE fix. TDD:
tests/unit/dast-method-not-allowed.test.ts gains the devices QUERY case (4/4).
Three latent heavy-CI reds surfaced by the VPS validation dispatch (the fast
path never runs these gates):
- t06 route-validation: POST /api/github-skills destructured request.json()
blind — a non-array 'targets' would .map-crash. Now validateBody(zod)
with defaults preserved (Hard Rule #7). Guard:
tests/unit/github-skills-route-validation.test.ts (4/4).
- env-doc-sync: document OMNIROUTE_SKIP_SYSTEM_TRUST (#6310) and the
changelog-integrity gate envs CHANGELOG_BASE_REF/ALLOW_CHANGELOG_REMOVALS
(#6300) in .env.example + ENVIRONMENT.md.
- zizmor ratchet: the new merge-integrity job's checkout/setup-node uses were
unpinned (+1 finding, 160 > baseline 159); pinned by SHA -> 158 (< baseline).
#6292 rewrote/added zero-width-marker tests with 7 new 'as any' result casts,
pushing the file past its frozen suppression (84) — and when violations
exceed the suppressed count ESLint reports ALL of them, so the ci.yml Lint
job went red with 90 errors. The 7 new sites get typed shapes (delta /
arguments / choices / output accessors — same pattern as fecf888fd); the
pre-existing debt stays frozen at the exact new count (83). 50/50 tests
green, file lint-clean under the suppressions baseline.
#6186 added omni-github-skills (API 22 -> 23) and updated computeCoverage's
total, but left the old count hardcoded in the A2A layer: listCapabilities
metadata reported coverage.api.total 22 (type literal + value) and
SkillCoverageSchema pinned z.literal(22) — so the schema would REJECT the
correct runtime value. Aligned all three to 23 + the unit fixtures
(listCapabilities-a2a, agentSkills-schemas, 46/46 with agentSkillTools-mcp).
