* fix(web-tools): anchor tool contract at prompt tail + user-turn reminder
The <tool> contract from prepareToolMessages was prepended as the first
system message. Web executors fold all system messages into one block, so
with agentic clients whose system prompts exceed ~28K chars the contract
sat at the head of a huge block and web models ignored it, refusing tool
calls with "tool X is not in my tool set" (chatgpt-web, 0/3 at 30K chars).
Two changes, both required in testing:
- Dual placement: the full contract now rides as a trailing system
message (folds to the tail of the system block) and a one-line
reminder naming the tools is appended to the latest user message.
- Rewording: the contract now frames injected tools as client tools
invoked via a plain-text protocol, distinct from the model's native
tool registry (web.run, python.exec, ...), and instructs the model to
never claim they are unavailable. Without this the model resolved
tool names against its native registry and refused even when it had
seen the contract.
Measured on cgpt-web gpt-5.5-thinking/gpt-5.6-thinking/o3: prepend 0/3
tool calls at 30K chars; dual placement 16/17 across 30K-250K system
prompts, 30-tool sets, multi-turn tool history, streaming, and 3-way
concurrency, with no spurious calls on no-tool prompts. Known limit:
~40K-char single user messages still flake (2/3) due to the upstream
model's own injection heuristics.
All prepareToolMessages consumers parse system messages
position-independently and select the current user turn by role scan,
so the trailing system message is shape-safe for every web executor.
* test(web-tools): cover contract placement edge cases
---------
Co-authored-by: Ryan Brosas <ryanjoserbrosas@gmail.com>
Wire Openference as a first-party OAuth gateway (PKCE, rotating refresh)
and an API-key catalog entry on api.openference.com, with live model
discovery, connection testing, free-tier badges, and regression tests.
Co-authored-by: Anh Tran <anhlead@outlook.com>
Every duckduckgo-web chat request failed with HTTP 418 ERR_CHALLENGE while
duck.ai worked normally in a browser from the same IP. Ground truth was
established by driving a real headful Chromium at duck.ai from that IP (it
returned 200), so the environment was never the problem — the anti-abuse
challenge solver was. Six independent defects were found; the first alone
disabled the solver completely.
1. Module syntax inside the vm sandbox source.
CHALLENGE_STUBS is executed with vm.runInContext, which compiles in SCRIPT
mode. A refactor mass-added `export` to the five `function` declarations
inside that template literal (they read as ordinary top-level TS functions),
so every solve threw SyntaxError. The executor swallows solve failures and
posts the raw unsolved challenge, which upstream answers with 418.
2. Double-escaped regex in a String.raw template.
`\\s` in __parseCssDisplay reached the sandbox as a literal backslash, so the
display regex never matched and a getComputedStyle probe silently read empty.
3. buildHtmlLookup undercounted descendants by one.
`count` backs el.querySelectorAll('*').length; that returns DESCENDANTS and
countHtmlElements already skips the #document-fragment root, so the `- 1` was
wrong. Chromium reports 3 for '<li><div></li><li></div'; we reported 2, and a
variant multiplies innerHTML.length by that count.
4. Browser-fidelity probes.
Newer challenge variants assert JS/DOM invariants a flat stub cannot satisfy:
real prototype chains (HTMLDivElement -> HTMLElement -> Element), NodeList
identity, a live body.children HTMLCollection, native-code toString, and
sloppy-mode `this === window`. Nine of thirteen failed. Notably Math must NOT
be sealed — Chromium reports Object.isSealed(Math) === false, and sealing it
made our vector differ by one.
5. The solved payload dropped meta.origin / meta.stack / meta.duration.
The duck.ai bundle always sends all three; captured browser requests confirm
it. Without them upstream returns 418 even when every client_hash is correct.
6. reasoningEffort is now mandatory on duckchat/v1/chat.
An otherwise byte-identical payload returns 200 with the field and 400
ERR_BAD_REQUEST without it (A/B verified live, repeated).
Also removes the throwaway "seed" chat POST that ran before every real request.
It existed to coax a usable challenge out of the upstream while the solver was
broken; it only doubled chat calls against an IP-rate-limited endpoint, showing
up as spurious 429 ERR_RATE_LIMIT.
Verification: the solver now reproduces real Chromium's probe vectors exactly
for all 8 captured challenge variants, and the executor returns 200 end-to-end
live (non-streaming, streaming, claude-haiku-4-5, and a math prompt returning
"42").
Tests: tests/unit/duckduckgo-challenge-solver-regression.test.ts (32 tests) and
tests/unit/duckduckgo-reasoning-effort-required.test.ts (5 tests), backed by
tests/fixtures/duckduckgo/challenge-variants.json — real captured challenge
programs plus the probe vectors a real browser produced for them, so the suite
asserts against recorded browser behaviour rather than our own output. Each fix
was confirmed to fail its test when individually reverted.
Co-authored-by: Mynacol <git@mynacol.xyz>
requestLogger.ts's cloneBoundedForLog had its own hardcoded depth cap of 6,
independent of the existing configurable getChatLogMaxDepth(). A typical
Chat Completions response body's responseBody.choices[0].message.tool_calls[0].function
sits at exactly depth 6, so every logged tool call's function field
(name+arguments) was silently replaced with the literal string "[MaxDepth]"
before ever being stored — corrupting the data, not just how it renders.
Bumped the shared default 6->20 and switched requestLogger.ts to read it
instead of using its own literal.
(cherry picked from commit a2df6cf289)
Co-authored-by: Markus Hartung <mail@hartmark.se>
* feat(logging): make the chat-log truncation limit configurable, bumped default 128x
The 8KB cap on logged request/response bodies
(open-sse/handlers/chatCore/logTruncation.ts::truncateForLog()) was
hardcoded — trivially exceeded by any real multi-turn agentic
conversation, meaning the dashboard's "Full Conversation" panel could
only ever show a placeholder instead of the actual messages for nearly
every logged row of any conversation with real substance.
- Added CHAT_LOG_MAX_BODY_KB env var (src/lib/logEnv.ts::
getChatLogMaxBodyBytes()), default 1024 KB (1MB) — a 128x bump from
the old hardcoded 8KB — following the same configurable-limit pattern
as the sibling CHAT_LOG_TEXT_LIMIT/CHAT_LOG_ARRAY_TAIL_ITEMS/etc. vars.
- Documented in .env.example and docs/reference/ENVIRONMENT.md.
estimateSizeFast() (open-sse/utils/estimateSize.ts) has been
substantially rewritten upstream since this bug was first found (now an
iterative Frame-based walker with a separate node-visit budget, not the
simple stack loop originally patched) — re-implemented the fix against
the current algorithm rather than porting the old diff: the byte
early-exit was unconditionally the module-level ESTIMATE_SIZE_BYTE_LIMIT
(256 KiB) with no way for a caller to raise it, so any caller comparing
against a bigger configured threshold could never see a size above
~256 KiB — every payload between 256 KiB and the caller's real limit
looked "under threshold" and truncation never fired, the opposite of
intended. Added an optional byteLimit parameter (default unchanged at
ESTIMATE_SIZE_BYTE_LIMIT, so isSmallEnoughForSemanticCache's existing
behavior is untouched) threaded through both the byte-check early-exit
and the node-budget-exhaustion fail-closed fallback, with
truncateForLog() now passing its own configured getChatLogMaxBodyBytes()
value through.
* feat(dashboard): show conversation session tag in request detail metadata
Adds a "Conversation" field to the request detail panel's metadata
grid (after "Combo"), showing the request's conversation id
(sessionTag) for quick reference/copy.
---------
Co-authored-by: Markus Hartung <mail@hartmark.se>
* fix(responses-api): sync reasoning-cache write index with the fixed read side
The turn-index-hardcoding fix updated the reasoning-cache read side
(translator/index.ts's main replay loop) to key lookups by the assistant
message's real position in the messages array, but two other spots still
used the old hardcoded convention:
- chatCore.ts's write side (both the streaming and non-streaming
completion paths) still cached every response under a hardcoded
messageIndex: 0.
- translator/index.ts's own plain-turn (non-tool-call) cache-key lookup
ALSO still hardcoded messageIndex 0 at its call site — a second,
previously undiscovered instance of the same class of bug, found while
re-verifying this fix against the current upstream tip (the original
fix only addressed the write side).
Past the first assistant turn these conventions no longer matched, so
DeepSeek/Xiaomi-mimo plain-turn reasoning replay silently missed the
cache and fell back to the placeholder (or, once #9573 removed the
placeholder fallback, to an absent field) in ordinary multi-turn
conversations.
Compute the write-side index from the incoming request's message count
instead, and use the real loop-provided messageIndex on the read-side
lookup, both matching the position the response occupies once the
client appends it to history for the next turn.
Note: this was originally part of a larger squashed fix (output_index
collision prevention across reasoning/message/tool_call items,
reasoning-content-alias generalization) that has since been superseded
by upstream's own independent fix — translator/response/openai-responses.ts
now has its own dense-output-index-sort + getReadableReasoningValue
implementation (own comment: "mirrors upstream PR #721"). Only this
narrower, still-genuinely-broken write/read index sync survives as a
distinct bug.
Test plan:
- TDD: tests/unit/reasoning-cache.test.ts's new end-to-end
"write side (chatCore's messageIndex) and read side (translateRequest)
agree on the same key end-to-end" test, plus the pre-existing
"should inject placeholder for a plain (non-tool-call) DeepSeek turn"
and "should replay cached reasoning for a plain (non-tool-call)
DeepSeek turn when available" tests — confirmed failing against the
pre-fix code on a clean release/v3.8.50 checkout (both the
hardcoded-0 write side AND the hardcoded-0 read-side lookup
independently reproduce the mismatch), passing after both fixes
- npm run typecheck:core — clean
- npm run lint — clean
- npm run check:file-size — clean (chatCore.ts rebaselined 5034->5042
for the messageIndex computation at both call sites;
reasoning-cache.test.ts frozen at 1035, matching the original fix's
own rebaseline)
- 2 pre-existing, unrelated test failures in the same file
("should replace empty-string reasoning_content with
NON_ANTHROPIC_THINKING_PLACEHOLDER on cache miss",
"should inject placeholder for a plain (non-tool-call) DeepSeek turn
missing reasoning_content") confirmed present on a completely clean,
untouched release/v3.8.50 checkout — these test obsolete
placeholder-injection behavior the code deliberately removed per
#9573 (see the code's own comment); not touched by this PR
* fix(chat): reduce file size
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
* fix(chat): reconcile file-size baseline
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
---------
Co-authored-by: Markus Hartung <mail@hartmark.se>
* fix(deps): bump transitive deps for 6 Dependabot + remaining audit vulns on main
Same overrides as #9464 (ip-address, hono, fast-uri, socket.io-parser, undici)
applied directly to main. Also covers brace-expansion (scoped), js-yaml v4 copies,
and mermaid.
npm audit: 6→0 vulnerabilities.
Closes Dependabot #161-#166.
* fix(deps): bump nanoid, dompurify for 2 new Dependabot alerts (#189, #190)
Bumps: nanoid ^3.3.17 (was transitive, now overridden), dompurify ^3.4.13
(with monaco-editor scoped override). Closes Dependabot #189, #190.
Remaining #182-#188 (js-yaml + mermaid) already closed by #9651 merge —
awaiting Dependabot re-scan.
npm audit → 0 vulnerabilities.
* fix(repo): harden .gitignore to also ignore a _tasks symlink (/_tasks)
_tasks is a SEPARATE nested git repo (gitignored). The pattern _tasks/ (trailing
slash) ignores only a directory, not a SYMLINK named _tasks. A self-referential
_tasks symlink can slip in via git add -A and, once pulled, checkout materializes
it over the real _tasks repo (destroying plans/specs/hands-off). Anchored /_tasks
ignores the symlink too, preventing re-capture.
* fix(translator): keep Responses namespace identity across the hub-and-spoke pivot
Step 1 of the pivot (openai-responses -> openai) flattens namespace sub-tools
to a qualified wire name (#8295) and records the `{namespace, name}` pair on a
non-enumerable `_toolNameMap`. Step 2 (openai -> target) returns a brand-new
object, so the property was dropped for every non-OpenAI target. chatCore then
handed `null` to the #7936 response seam and namespace sub-tool calls reached
the client under their flattened name, which Codex rejects with
`unsupported call: <name>` — the symptom #7936 was opened to fix.
Copying `_toolNameMap` through is not viable: openai-to-claude and
openai-to-gemini publish their own `Map<string, string>` alias map on that same
property during step 2, so it carries two incompatible types. This adds a
dedicated `_namespaceToolIdentityMap`, propagated by translateRequest across
the pivot; chatCore prefers it and falls back to `_toolNameMap` for the
non-pivot producers. Both keys are stripped from the cliproxyapi wire body.
Fixes#9780
* fix(chat): reduce file size
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
* fix(chat): reduce combined file size
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
* fix(chat): reduce combined file size
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
---------
Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com>
Co-authored-by: VXNCXNX <vincent@preuve.ai>
* fix(sse): apply Azure request-param rules on the azure-ai wire path
Azure rejects several stock Chat Completions params on its newer deployments
and returns HTTP 400 rather than ignoring them:
max_tokens -> 'max_tokens' is not supported with this model.
Use 'max_completion_tokens' instead.
reasoning_effort -> Function tools with reasoning_effort are not supported.
Those rules lived inline in AzureOpenAIExecutor, so they only covered the
azure-openai provider. azure-ai (Azure AI Foundry) had no executor entry and
fell through to the bare DefaultExecutor, so the SAME Azure deployment
succeeded on one connection and 400'd on the other. Every agentic client sends
tools on every turn, so azure-ai failed on the first request.
Extract the rules to open-sse/executors/azureParamRules.ts, add an
AzureAiExecutor that inherits DefaultExecutor's azure-ai URL/header/apiType
handling unchanged and applies the shared rules, and register it for azure-ai.
Also widen the deployment pattern to cover gpt-chat-latest: it is a moving
alias that resolves to a GPT-5-era model and rejects max_tokens, but carries no
version number for the token-boundary pattern to key on. Verified against the
base regex - gpt-chat-latest did not match, which is exactly the observed 400.
Regression guard: tests/unit/azure-param-rules.test.ts, including an assertion
that getExecutor("azure-ai") no longer resolves to a bare DefaultExecutor.
* fix(sse): clamp Azure gpt-4o-mini completion tokens to its 16384 ceiling
Azure gpt-4o-mini deployments accept at most 16384 completion tokens and 400 on
anything larger:
max_tokens is too large: 32000. This model supports at most 16384 completion
tokens, whereas you provided 32000.
The 32000 is OmniRoute's own doing: adjustMaxTokens raises any smaller
max_tokens to DEFAULT_MIN_TOKENS (32000) whenever tools are present, to avoid
truncated tool arguments. That floor has no upper bound, so an agentic client
asking for far less still trips the model ceiling on its first turn.
Add scoped maxOutputCap rules in paramSupport.ts for both Azure wire paths.
PROVIDER_MAX_TOKENS is the wrong lever here - it is provider-wide, and the same
Azure resource also serves GPT-5 deployments with a much higher ceiling.
Regression guard: tests/unit/azure-max-output-clamp.test.ts, which also pins
that the clamp does not leak to gpt-5.1 or to gpt-4o-mini on other providers.
---------
Co-authored-by: Mihaly Bodo <michael@proton-quantum.com>
copyOpenAICompatibleReasoningFields only stripped the sentinel
(NON_ANTHROPIC_THINKING_PLACEHOLDER = "(prior reasoning summary
unavailable)") from reasoning_content and reasoning. Non-standard
reasoning fields (reasoning_text, thinking, thought) and
reasoning_details items passed through raw, leaking the internal
replay sentinel to clients on providers that use those fields
(e.g. Venice), where the model echo surfaces as a bogus thought block
and can degrade into empty turns.
Strip the sentinel from every forwarded reasoning field, including
per-item text/content inside reasoning_details; drop items/fields that
strip to nothing while preserving non-text details such as
reasoning.encrypted.
Fixes#9765
Refs #8081, #9606
Co-authored-by: safeer <asafeer1994@gmail.com>
Live incident (2026-08-08): an OpenClaw agent sent a short preamble line
("Kör nu, på riktigt — apply_patch på vibe-scriptet:") followed by an
apply_patch tool call in the same turn. The client only spoke the preamble
and never executed the patch, even though OmniRoute's own recorded
responseBody had a complete, valid tool_calls entry.
Root cause: emitToolCall/closeToolCall computed a tool call's output_index
as `reasoningIndex + 1 + tcIdx`, assuming reasoningIndex + 1 was free for
the first tool call (tcIdx=0). But a text message emitted in the same turn
ALSO claims reasoningIndex + 1 (or index 0 with no reasoning) — so a
turn with reasoning + text content + a tool call collided the tool call's
added/delta/done events onto the same output_index as the just-closed
message. A client that tracks response items by output_index (as expected
for the Responses API) sees the tool call events land on an index it
already marked complete and can silently drop them.
Fix: track whether a message item was actually emitted at that index
(state.msgItemAdded) and, if so, tool calls start one slot after it.
Extracted a shared toolCallOutputIndexBase() helper so emitToolCall and
closeToolCall can no longer compute this independently and drift apart.
Confirmed via the live call log artifact (id 1786223153235-770a1c):
response.output_item.done for the text message and response.output_item.added
for the tool call both carried output_index=1 in the raw SSE stream, 1.84s
apart, exactly matching the reported symptom.
Co-authored-by: Markus Hartung <mail@hartmark.se>
* fix(executors): strip redundant oneOf matching sibling enum
The Codex private Responses endpoint intermittently returns a 502 upstream_empty_response for tool parameters that combine oneOf:[{const,...}] with a sibling enum containing the same value set.
When the const and enum sets match exactly, oneOf adds no constraint beyond enum. Add stripRedundantOneOfConstEnum to normalizeCodexTools to remove only this semantically redundant form.
The schema-aware recursive walker requires non-empty, unique string const branches containing annotations only, string enum values, and an exact set match. It preserves bare oneOf[const], narrowing or non-matching sets, type-discriminated oneOf, empty oneOf, non-string values, and anyOf/allOf.
Run the normalization after stripUnsupportedRegexPatterns and before assigning tool.parameters. Add focused regression coverage for matching, non-matching, nested, immutable, and Chat-to-Responses cases.
* docs(changelog): update PR number in changelog fragment
---------
Co-authored-by: Vasily Larin <larin.vas@outlook.com>
* fix(cursor): hydrate SelectedImage via blobIdWithData + JPEG soft-cap
Cursor vision expects SelectedImage.blob_id_with_data (field 9) backed by
the session blobStore, and large clipboard PNGs need JPEG soft-cap prep
rather than a hard 1 MiB reject before encode.
* docs(changelog): add fragment for Cursor SelectedImage blobIdWithData fix
* refactor(cursor): split image protobuf encoding
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
---------
Co-authored-by: SB Yoon <44089734+yansigit@users.noreply.github.com>
isStreamingUpstreamError used a key-presence check (parsed.error != null)
which false-positives on benign values some backends emit on every chunk
({}, '', false, 0). When opencode issues a tool-call turn, the upstream SSE
opens with role-only frames (no recognized content) and a later chunk that
carries real tool_calls content PLUS a benign empty error field. The error
gate runs BEFORE content recognizers, so that single frame short-circuits
to 'error' -> 502 'streaming upstream error'. Same combo via kilocode works
because its wire format never emits the empty error field.
Fix: isSubstantiveError() helper — only treat error as real when it carries
non-empty string, non-empty object, or explicit true. Empty object {}, empty
string '', false, and 0 are benign.
TDD: tests/unit/quality-validation-benign-error.test.ts proves tool_calls
chunk with error:{} or error:'' is valid (was 502), while a real error
{message, code} still correctly fails.
* fix(ci): drop unused RadarReferrals type export — dead-code ratchet back to 227 baseline
The radar referral-links feature (#9697) exported the inferred type
RadarReferrals from feedSchema.ts but nothing imports it (the singular
RadarReferral is the consumed type). knip counts it as a new dead export,
pushing the dead-code ratchet to 228 > 227 and failing Fast Quality Gates
on every PR born after the merge. RadarReferralsSchema itself stays — it
is used by RadarFeedSchema.
Refs #9737
* fix(ci): clear the 08-08 base-red layer — prod crash in chat.ts, Responses API payload regression, born-red stdio test, gate drifts
Six independent base-reds from the 08-07 evening merge batch, each verified
against the pure release/v3.8.50 tip:
- src/sse/handlers/chat.ts: #9467's squash carried a refactor hunk that
renamed the all-rate-limited breaker guard to an UNDEFINED variable
(isAllRateLimited) — a production ReferenceError on the all-accounts-429
path (chat.ts is outside typecheck:core scope, so only tests caught it).
Restore credentials?.allRateLimited. Guard: chat-rate-limit-body-lock (2/2),
also un-breaks batch_api and chat-combo-live-test.
- open-sse/utils/stream.ts: #9315 switched providerPayload summaries to the
accumulated responseBody, but in passthrough paths that body is synthesized
in chat-completion shape — Responses API lost its `response` object in the
dashboard payload. Keep the events-derived summary for OPENAI_RESPONSES
only. Guard: stream-utils + stream-collector-9315 suites (51/51).
- tests/unit/mcp-stdio-json-purity.test.ts: born red — the full CLI chain
takes ~10s (2x tsx import + DB init) and the test slept a fixed 4s. Poll
for the first stdout line with a 60s deadline instead.
- tests/unit/plugins-route-error-sanitization.test.ts: register #9445's new
marketplace/install route in PLUGIN_ROUTES (route already sanitizes) (33/33).
- tests/unit/provider-models-route-codex.test.ts: realign pinned GPT-5.6
input limit to #9432's deliberate 272000→922000 bump (7/7).
- lint: fix 11 no-explicit-any errors in repro-9630 + specialty-9293 tests,
prune 1 orphaned suppression, allowlist the opencode-ai devDependency
(#8869, publisher-verified), and reword a doc line the fabricated-docs
gate misread as an env var.
Gates re-verified locally: lint:json --max-warnings 0 exit 0, dead-code 227,
typecheck:core clean, check:deps OK, check:fabricated-docs OK.
Refs #9737
* fix(ci): clear the third 08-08 base-red layer — invalid ru rule pack, stale event pin, orphaned UI repro test, pack/mutation/file-size drifts
Follow-up to the previous layer: the serial fast-gates chain unmasked one
more stratum after file-size/dead-code went green, all verified against the
merged release/v3.8.50 tip:
- compression rules ru/ultra.json (#9581): two rules shipped
minIntensity "notes", which is not a valid CavemanIntensity
(lite|full|ultra) — loading ANY language pack list threw and killed the
rtk-loader suite. Mapped both to "ultra" (they are the most aggressive
punctuation/case rules, matching the en pack tiers). 2/2.
- plugins-welcome-banner-e2e: #9668 added the onStreamComplete builtin
event (real emission path via runOnStreamCompleteHooks) and missed this
pinned-list sibling. 35/35.
- tests/unit/free-pool-frontend-repro (#9046): landed as .tsx with
node:test semantics — no runner collects tests/unit/*.tsx, so it NEVER
ran (test-discovery NEW-orphan). It contains zero JSX; renamed to .test.ts
so the unit runner's existing glob collects it. 5/5 (first real run).
- pack-policy: allow + require bin/mcpStdioConsoleGuard.mjs (#9281) — it is
preloaded via node --import by bin/mcp-server.mjs, so a published artifact
without it crashes 'omniroute --mcp' at startup.
- stryker.conf.json: add 5 covering unit tests from the batch (#8779/#9204/
#9330/#9630/openrouter-passthrough) to tap.testFiles (--strict drift).
- file-size-baseline: consolidate the base-drift rebaseline for the 12
files grown by the 08-06..08-08 batches (#9616's entries never reached the
base; measured on this branch's tree — this PR's own source edits add zero
lines to any frozen file).
Local battery: file-size/deps/test-discovery/mutation/pack-policy/dead-code/
duplication/docs-all/secrets/vuln/workflows ratchets all exit 0; full lint
gate --max-warnings 0 exit 0.
Refs #9737
* fix(types): clear the 3 uncovered open-sse-typecheck regressions + realign combo skip-code siblings
Fourth base-red layer unmasked by the serial gates. The other 4 typecheck
regressions (codex.ts, kiro.ts, tierResolver.test.ts, translator/index.ts)
already have dedicated open [TS7] PRs (#9748/#9753/#9742/#9747) — not
duplicated here. This commit covers only what no open PR owns:
- devin-agentic/serializer.ts TS2367: drop the dead 'role === "system"'
branch — the guard above already narrows role to user|assistant (system
throws unsupported_role). Devin suites 104/104.
- raycast.ts TS2416: the buildHeaders 'override' never matched the base
signature (2nd param is the signed payload string, not the stream
boolean) — renamed to a private buildRaycastRequestHeaders helper so a
polymorphic buildHeaders(credentials, true) call can never bind here.
- modelMetadataRegistry.ts TS2352: PricingByProvider → nested-record cast
now goes through unknown (shape is runtime-guarded by findInsensitive).
- combo-routing-engine.test.ts: realign 2 pre-dispatch-skip expectations to
#9630's deliberate ALL_TARGETS_SKIPPED contract (87/87).
Refs #9737
* fix(ci): clear the fifth 08-08 base-red layer — reasoning-placeholder contract sweep, GPT-5.6 limits sweep, vi key parity
The 08-08 merges (#9610 reasoning replay, #9432 GPT-5.6 limits, #9630 combo
skip codes, #9336 provider key links) each changed a contract and left
sibling tests pinning the old one. Full grep sweep per contract, not just
the shard that happened to go red:
- reasoning placeholder (#9573/#9610): the fix DELIBERATELY removed
NON_ANTHROPIC_THINKING_PLACEHOLDER injection on cache miss — the model
echoed the placeholder as its own reasoning (empty stop) and re-poisoned
cache + client history; DeepSeek's 400 is specific to an EMPTY STRING, not
an absent field. Realigned reasoning-cache (2 cases, renamed to describe
omission) + tool-request-sanitization (1 case + dead import). 60/60.
- GPT-5.6 Codex limits (#9432, 272000 -> 1050000 ctx / 922000 input):
realigned vscode-token-routes-gpt56 (2) + vscode-token-routes (3). 43/43
together with t23-t24.
- combo skip codes (#9630): t23-t24-fallback-resilience T24 now expects
ALL_TARGETS_SKIPPED like the combo-routing-engine siblings.
- vi.json key parity: #9336 added providers.getApiKey/getApiKeyDescription
to en.json without syncing vi (the only locale with a parity gate).
Translated both; providers block reordered to match en key order. 5/5.
- pack-artifact-policy.test.ts: sibling of this PR's own required-paths
change (bin/mcpStdioConsoleGuard.mjs). 10/10.
- combo-routing-engine.test.ts: dropped the 6 comment lines added in the
previous commit so the frozen test file-size stays at its baseline (the
rationale lives in that commit message, not the test body).
Gates: file-size, test-discovery, mutation-test-coverage, pack-policy,
open-sse-typecheck, dead-code all exit 0.
Refs #9737
* fix(translator): keep the reasoning_content placeholder for Xiaomi MiMo — #9610 traded one live 400 for another
The xiaomi-mimo replay test (9router#1321) went red on the base after #9610
removed the NON_ANTHROPIC_THINKING_PLACEHOLDER injection globally. That test
is NOT stale — it guards a documented upstream 400 ('Param Incorrect: The
reasoning_content in the thinking mode must be passed back to the API'), so
realigning it would have masked a reintroduced production bug.
Two real bugs conflict here:
- #9573: forwarding the placeholder makes the model continue its chain of
thought FROM that text (echo -> empty stop) and re-poisons cache/history.
- 9router#1321/#1337: omitting reasoning_content on a plain replay turn makes
Xiaomi MiMo reject the request outright.
#9610's evidence for omitting is provider-specific — it verified that
deepseek-v4-flash accepts an ABSENT field. It does not extend to MiMo. So the
omission stays for every provider #9610 covered, and the placeholder survives
the cache miss only for xiaomi-mimo (new requiresReasoningContentPresence
predicate next to isReasoningOnlyReplayTarget). The echo that comes back is
still stripped on the way in by isInternalReasoningPlaceholder(), so #9573's
cache/history poisoning stays fixed for MiMo too.
Both contracts now hold simultaneously: xiaomi-mimo replay + reasoning-cache +
tool-request-sanitization 61/61; placeholder-strip/responses/translator/combo
regression sweep 168/168. Gates: file-size, open-sse-typecheck, dead-code,
mutation-test-coverage exit 0; typecheck:core clean.
A live check on the VPS (Hard Rule #18 path 2) is the only way to confirm the
DeepSeek half of #9610's empirical claim; flagging it in the PR rather than
widening this fix on speculation.
Refs #9737
* test(translator): pin the reasoning-placeholder provider scope so neither half of the conflict can silently re-break
#9610 removed the placeholder globally on the strength of ONE provider's
observed behavior (deepseek-v4-flash accepting an absent reasoning_content),
which re-opened the MiMo 400 (9router#1321). The previous commit scoped the
placeholder to xiaomi-mimo; this pins BOTH directions in one test so the next
global edit fails loudly instead of trading the bugs again:
- xiaomi-mimo plain replay turn, cache miss -> reasoning_content present
(narrowing the scope away from MiMo re-opens 9router#1321)
- deepseek plain replay turn, cache miss -> reasoning_content absent
(widening it back to DeepSeek re-opens the #9573 echo bug)
Guard verified by mutation: forcing requiresReasoningContentPresence() to
return true makes the DeepSeek half fail (1 pass / 1 fail), and the file was
restored from the pre-probe copy before committing.
Also checked kimi-coding/kimi-coding-apikey, the other strict-contract entries
in REASONING_REPLAY_PROVIDERS: their originating PR (#7673) fixes capture and
replay of REAL reasoning and documents no 400 on an absent field, so they stay
out of the placeholder scope — evidence-scoped, not speculatively widened.
Reasoning suites together: 87/87. Gates: file-size, test-discovery,
mutation-test-coverage, dead-code exit 0; eslint clean.
Refs #9737
---------
Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com>
