mirror of
https://github.com/diegosouzapw/OmniRoute.git
synced 2026-07-31 04:12:10 +03:00
feat/plugin-browser-pool
262 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
Rafael Dias Zendron
|
7322740e7e |
fix(#8141): log pending request counter decrement failures (#8179)
* fix(ci): resolve upstream-inherited check failures * fix(streamHandler): log trackPendingRequest decrement failures instead of swallowing The clearPendingRequest function had an empty catch block around the trackPendingRequest decrement call. If it threw, the pending request counter stayed incremented — causing drift, false-positive rate limiting, and masked overload conditions. Now logs the error with context for observability. Fixes #8141 --------- Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
小妍儿 ✨
|
b640b69078 |
feat(codex): support reference image edits (#8122)
* feat(codex): support reference image edits * docs(changelog): add Codex edit fragment * fix(codex): harden image edit admission * fix(security): redact image error credentials * fix(security): close error redaction bypasses * feat(codex): support multiple image references * fix(codex): preserve reference candidate semantics * chore(quality): rebaseline image-generation-handler.test.ts for #8122 codex image edits own-growth Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: 千乘妍 (Xiaoyaner) <xiaoyaner0201@users.noreply.github.com> Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Markus Hartung
|
cc17b304ab |
fix(sse): Gemini TPM/RPD quota classification + combo cooldown-wait resilience (#8213)
* fix(sse): Gemini TPM classification, combo-cooldown-wait for auto/quota-share, and target-timeout floor
Gemini TPM/RPM 429s were misclassified as QUOTA_EXHAUSTED because
sanitizeErrorMessage() truncates to the first line, hiding Google's
metric name and retry hint on lines 2-3. Added a rawMessage field
(internal-only, never reaches the client) and classifyGeminiQuotaMetricFromText()
to classify from the untruncated text, reordered ahead of the generic
credits/daily-quota checks.
Widened comboCooldownWaitEnabled (wait out a short transient cooldown
instead of crystallizing a 429/503) from quota-share-only to also cover
auto-strategy combos, and raised the wait ceiling to 65s/130s-budget/90s-cap
to match Gemini's ~60s TPM/RPM windows.
The per-target timeout (DEFAULT_COMBO_TARGET_TIMEOUT_MS, 120s) was shorter
than the new 130s cooldown-wait budget, so a target could get cut off
mid-wait with a synthetic 524 instead of completing the retry. Added
resolveComboTargetTimeoutMsForCombo()/isComboCooldownWaitEligible() in
comboConfig.ts to raise the per-target floor to budgetMs+buffer only for
wait-eligible strategies (auto/quota-share), verified live: a 12-request
concurrent burst against a TPM-exhausted combo went from 2/12 succeeding
(10 x 524) to 12/12 succeeding with zero 503/524.
Also: liveGeminiShared.ts's sendAndValidate now fails fast on a 503
instead of retrying past it, and the health dashboard + request logger
surface TPM stats alongside RPM/RPD.
Co-authored-by: Markus Hartung <markus.hartung@gmail.com>
* fix(sse): combo-exhausted rejection logs now capture request body + attempted models
recordRejectedRequestUsage() (the fast path for combo requests that never
reach handleChatCore, e.g. all targets locked by resilience cooldown)
hardcoded provider: "-" and never passed a request body to saveCallLog(),
so /dashboard/logs entries for these failures were nearly useless for
debugging: no way to see the client's request or which models were tried.
- recordRejectedRequestUsage() now accepts requestBody and persists it
through the existing saveCallLog() artifact mechanism (same path
handleChatCore's own logging uses).
- Added summarizeComboAttemptedModels(), which reads the combo's own model
list (always available, unlike the response's combo-diagnostics headers —
a model-level resilience-lockout skip never touches the
exhaustedProviders/exhaustedConnections sets those headers are built
from) to populate a real "provider" value instead of "-".
- Wired both into the call site in src/sse/handlers/chat.ts.
NOTE: unrelated to the Gemini TPM/combo-cooldown-wait fix on this branch —
landed here per operator request, to be split into its own branch/PR.
Co-authored-by: Markus Hartung <markus.hartung@gmail.com>
* feat(sse): synthetic streaming keep-alive event + 5-minute Gemini cooldown-wait ceiling
Many clients enforce a first-SSE-byte timeout, which made it unsafe to wait
out a longer upstream rate-limit cooldown on a streaming request — the
client would abandon the connection before any bytes arrived. This landed
in two parts:
1. Synthetic startup "thinking" event (OpenAI chat/completions format):
the already-existing withEarlyStreamKeepalive wrapper (open-sse/utils/
earlyStreamKeepalive.ts, wired into /v1/chat/completions, /v1/messages,
/v1/responses since #2544) opens the SSE stream immediately once a
request runs past its threshold, but only ever sent empty/no-op
keepalive frames. Added a `startupFrame` option (defaults to
`keepaliveFrame` — zero behavior change unless a route opts in) so the
very first frame can carry real content instead. Wired
OPENAI_STARTUP_THINKING_FRAME (a reasoning_content delta: "OmniRoute:
got request, sending to provider") into /v1/chat/completions only —
Claude Messages and Responses API formats both require a preceding
envelope event (message_start / response.created) that a synthetic
pre-dispatch frame can't safely fabricate without risking a duplicate
envelope once the real stream arrives, so those two routes keep their
existing (safe, proven) keepalive frames unchanged.
2. Raised the "wait out a known cooldown, then retry" ceiling to 5 minutes
for both retry mechanisms, now that a client-side first-byte timeout is
no longer a risk on the (opted-in) route:
- comboCooldownWait (auto/quota-share combos, open-sse/services/combo.ts):
maxWaitMs hard clamp raised 90s -> 300s (src/lib/resilience/settings/
normalize.ts); defaults raised to maxWaitMs:90s/maxAttempts:5/
budgetMs:300s. comboConfig.ts's resolveComboTargetTimeoutMsForCombo
already derives the per-target timeout floor from budgetMs, so it
tracks the new ceiling with no further changes.
- waitForCooldown (direct, non-combo model requests, src/sse/handlers/
chat.ts): this mechanism had NO cumulative cap before — only a
per-wait cap (maxRetryWaitMs) and a retry count (maxRetries), so
maxRetries x maxRetryWaitMs could exceed 5 minutes with no ceiling.
Added a budgetMs field (mirrors comboCooldownWait) to
WaitForCooldownSettings/CooldownAwareRetrySettings, threaded a
requestRetryBudgetLeftMs tracker through chat.ts's requestAttemptLoop
(mirrors combo.ts's comboCooldownBudgetLeftMs), and made
getCooldownAwareRetryDecision refuse to wait once the cumulative
budget is exhausted even if the single wait is under maxRetryWaitMs.
Co-authored-by: Markus Hartung <markus.hartung@gmail.com>
* fix(sse): extend the synthetic keep-alive thinking event to /v1/responses
Live incident (OpenClaw, log id 1784407081908-cbc24f): a /v1/responses
request to gemini/gemma-4-31b-it took 56s to produce a first byte and the
client disconnected (499 request_signal_aborted) — the same client-first-byte-
timeout problem the previous commit fixed for /v1/chat/completions, but
/v1/responses only had the generic bare-comment keepalive (no content), so it
wasn't covered.
Added RESPONSES_STARTUP_THINKING_FRAME: a self-contained synthetic reasoning
item (response.output_item.added -> reasoning_summary_part.added ->
reasoning_summary_text.delta -> reasoning_summary_part.done), opened AND
closed within this one frame rather than left dangling — it never carries a
response_id, so it can't collide with the real upstream response's own
independent response.created lifecycle that follows. Mirrors the abbreviated
delta+part.done close pattern open-sse/utils/stream.ts's own
emitSyntheticResponsesReasoningSummary already uses for real mid-stream
reasoning content.
Wired into src/app/api/v1/responses/route.ts via the startupFrame option
added in the previous commit.
Co-authored-by: Markus Hartung <markus.hartung@gmail.com>
* fix(sse): combo cooldown-wait vars reset every setTry, crystallizing a bogus 503 instead of waiting
Live incident (log id 1784416706646-51): a request to the "default" combo
(strategy=auto, maxSetRetries=3) hit a real Gemini TPM 429 on both gemma-4
targets, correctly classified as a short 40s rate_limit lockout — then
crystallized a 503 "all upstream accounts are inactive" in 6.9s instead of
ever reaching the cooldown-aware wait.
Root cause: `lastError`/`earliestRetryAfter`/`lastStatus` were declared with
`let` INSIDE the `for (setTry...)` loop body, so they reset to null at the
start of every set-try. When both targets lock out on setTry 0, every
subsequent setTry (1..maxSetRetries) pre-skips both targets via the
isModelLocked check with no real dispatch — so on the FINAL setTry (the only
one whose values the post-loop decision reads, since it's gated behind
`if (setTry < maxSetRetries) continue`), lastStatus was null, hitting the
"!lastStatus" branch (ALL_ACCOUNTS_INACTIVE 503) and completely bypassing the
comboCooldownWaitEnabled / earliestRetryAfter wait logic — even though a
real 429 with a known ~40s retry-after WAS observed on setTry 0.
This bug predates today's Gemini TPM work (any combo with maxSetRetries > 0
whose targets all lock out on the first pass was affected) but was masked in
existing tests: the "auto strategy (2 models...)" regression test uses
maxSetRetries: 0, so it only ever runs ONE setTry iteration and never
exercises the reset-on-retry path. It also explains why the dedicated
12-concurrent-request burst test passed cleanly — with concurrent requests,
timing variance meant some request's FINAL setTry iteration still had a live
target to dispatch to, giving lastStatus/earliestRetryAfter fresh data. A
single isolated request has no such luck.
Fix: hoist lastError/earliestRetryAfter/lastStatus to just inside
dispatchWithCooldownRetry, before the setTry loop, so they persist across
set-tries (still reset fresh on each recursive dispatchWithCooldownRetry()
call after a wait, which is correct). recordedAttempts/fallbackCount/
exhaustedProviders etc. are intentionally left per-iteration (unrelated to
this bug).
New regression test in tests/unit/combo-quota-share-cooldown-wait.test.ts
reproduces the exact live scenario (2 targets, both lock out on setTry 0,
maxSetRetries: 3) — confirmed red (503) against the pre-fix code, green
(200, waits and retries) against the fix.
Co-authored-by: Markus Hartung <markus.hartung@gmail.com>
* test(sse): extend live Gemini workload to Responses API + add large-context TPM test
Two additions to the live Gemini test suite, both live-verified against the
dev instance:
1. sendAndValidate() (tests/integration/liveGeminiShared.ts) now accepts an
apiFormat: "chat" | "responses" parameter, building the Responses-API
request shape (input array, max_output_tokens) and parsing its SSE events
(response.output_text.delta / response.reasoning_summary_text.delta /
response.completed) via the new readResponsesSSEStream(). Wired into two
new tests in live-gemini-workload.test.ts ([30]/[31]), mirroring the
existing Chat Completions streaming coverage. Verified live: 24/25 + 5/5
payloads succeeded end-to-end through the new code path (the one failure
was a ~300s test-client fetch timeout unrelated to the Responses API code
itself — a separate, not-yet-addressed test-harness limitation).
2. genHugeContextMessage() builds a single message large enough (~4
chars/token estimate) to approach or exceed Gemini's free-tier TPM ceiling
(16000 input tokens/min for gemma-4) by itself. Every other prompt
generator in this file tops out around 1-2k tokens — nowhere near that
ceiling — so none of the existing workload tests ever exercised a REAL TPM
429, only RPM-style rate limiting. tests/integration/gemini-large-context-tpm.test.ts
sends two ~12-13k-token requests back-to-back (comfortably exceeding
16000/min together) to exercise the full path against production Gemini:
TPM classification, the comboCooldownWait retry, and the synthetic
keep-alive frame on a genuinely slow request.
Co-authored-by: Markus Hartung <markus.hartung@gmail.com>
* fix(sse): abandoned combo target dispatch now observes its own per-target timeout, fixing a permanent "pending" dashboard leak
Live incident (dashboard log id 1784418258231-14961a, reported as "an ongoing
request even though there's already a 200"): a combo target dispatch
abandoned by comboTargetTimeoutMs (open-sse/services/combo/targetTimeoutRunner.ts)
left a permanent phantom "pending" entry in the dashboard, even after the
overall combo request had already succeeded via a different retry.
Root cause: chatCore.ts's createStreamController — and everything downstream
that depends on it (withRateLimit's Promise.race against Bottleneck,
acquireAccountSemaphore) — only ever watches clientRawRequest.signal, which
is the ORIGINAL client's request signal (set once via buildClientRawRequest
and reused unchanged across every target dispatch in a combo). It has no
connection to targetTimeoutRunner.ts's OWN AbortController
(target.modelAbortSignal), which is what actually fires when
comboTargetTimeoutMs (300s) elapses. src/sse/handlers/chat.ts's
handleSingleModel bridge between combo.ts and handleSingleModelChat received
`target.modelAbortSignal` but silently dropped it — never forwarded it
anywhere. So when a target got abandoned (e.g. stuck inside a wedged
Bottleneck rate-limiter queue, see the WEDGED force-reset log line from the
same incident), its per-target timeout fired and let the COMBO move on and
retry successfully elsewhere — but the abandoned dispatch's own promise
chain never learned it had been superseded, so it hung forever waiting on a
signal that was never going to fire, and trackPendingRequest(false) (the
finalize call) never ran.
Fix: thread target.modelAbortSignal through as a new modelAbortSignal
runtimeOption, and merge it into clientRawRequest.signal (via the existing
mergeAbortSignals helper from open-sse/executors/base.ts) right before
dispatch, so an abandoned target's own promise chain now observes its abort
and can reach its cleanup path — new resolveDispatchClientRawRequest() makes
this mechanically testable in isolation.
Co-authored-by: Markus Hartung <markus.hartung@gmail.com>
* fix(sse): combo cooldown-wait state recording, rate-limit wedge recovery, OpenAI-format SSE error frames
Five related fixes surfaced by live incidents (dashboard log ids 1784457764961-73,
1784465227489-a2cbc0, 1784504040241-6f8b9a) while validating the Gemini TPM/cooldown-wait
work on this branch against real OpenClaw traffic:
- combo.ts: the model-lockout bail-out branches in dispatchWithCooldownRetry never
recorded lastStatus, so once every target in a set hit an existing lockout the final
check crystallized a bogus ALL_ACCOUNTS_INACTIVE 503 instead of reaching the
cooldown-wait decision, even with a real 429 + short retry-after observed.
- combo.ts/combo/types.ts: the "all credentials cooling down" pre-dispatch rejection
(buildModelCooldownBody) nests its retry hint as error.retry_after/reset_seconds, not
the top-level retryAfter every other 429 shape uses — combo's extraction only read the
latter, so earliestRetryAfter stayed null for this shape even after lastStatus was fixed.
- rateLimitManager.ts: the wedge-recovery watchdog used disconnect(), which releases the
heartbeat timer but never rejects jobs already QUEUED on that instance — orphaned
dispatches hung until the outer ~300s per-target timeout, well past real clients'
patience. Switched to stop({ dropWaitingJobs: true }), safe because the wedge condition
already requires RUNNING===0 && EXECUTING===0.
- earlyStreamKeepalive.ts: the in-band error frame emitted after committing to a 200 SSE
stream was hardcoded to Anthropic's `event: error` convention for every route, including
the OpenAI-format ones (/v1/chat/completions, /v1/responses) where that framing is
either invisible or malformed to a plain data-line parser. Added per-route
OPENAI_CHAT_ERROR_FRAME / OPENAI_RESPONSES_ERROR_FRAME and wired them in.
- chatCore.ts: persisted a synthetic clientResponse error body even when the client had
already disconnected (AbortError) before that body was ever computed — misleading the
dashboard into showing "what the client received" for a response that was never sent.
Also: RequestLoggerDetail.tsx — Provider/Client Event Stream panes lost their collapse
toggle when StreamSection replaced the collapsible PayloadSection (
|
||
|
Markus Hartung
|
4ea08f520b |
fix(sse): Gemini malformed function-call handling + tool_choice translation (#8211)
* fix(sse): synthesize tool_calls for Gemini's malformed function-call abort reasons Live incident (dashboard log id 1784489701456-d8c0e9): Gemini terminates a stream with finishReason MALFORMED_FUNCTION_CALL/UNEXPECTED_TOOL_CALL when its own parser rejects an attempted tool call — there's no real functionCall part, only a human-readable finishMessage. gemini-to-openai.ts passed this through raw as finish_reason (9router#2462's fix, correctly keeping it off a clean "stop"/Claude end_turn), but a raw "malformed_function_call" isn't one of OpenAI's 5 documented finish_reason values, so a real OpenAI-format client (OpenClaw) has no handling for it at all and silently never notices the turn failed — confirmed live via tests/integration/live-gemini-workload.test.ts's [28] streaming case after the Gemini TPM/rebase work on this branch. Fix: synthesize a tool_calls entry (arguments carry the error code + Gemini's finishMessage, valid JSON) and finish_reason: "tool_calls" instead, routing the failure into the ordinary "tool call arguments didn't parse" path every OpenAI-compatible agent loop already handles. Defers to a real tool call if one already completed earlier in the same turn — the real call wins, no synthetic entry piles on top of it. Tests (TDD, each confirmed red-before-green): - 5 new unit tests in the existing 9router#2462 regression file, covering the synthesis itself, UNEXPECTED_TOOL_CALL, the real-tool-call-wins edge case, and no-regression on a clean STOP. - New fixture (tests/fixtures/translation/gemini-malformed-function-call-stream.json): the real 6-chunk event series from the live incident, sanitized (personal paths/URLs replaced with generic placeholders, structure preserved exactly). - New integration test chains the real translator into the real Responses API transformer using that same fixture, proving correct behavior on BOTH /v1/chat/completions and /v1/responses from one shared ground-truth event series. Co-authored-by: Markus Hartung <markus.hartung@gmail.com> * fix(sse): don't drop a malformed tool-call failure when it lands beside a real one Live incident (dashboard log id 1784589106014-2a42f8), analyzing why the prior malformed-function-call fix (3568c7259) still wasn't reaching the client in this case: Gemini can emit a REAL, valid functionCall AND finish the SAME candidate with MALFORMED_FUNCTION_CALL — the model attempted multiple tool calls in one turn (here: a real status-check call plus a malformed "exec"+"cron" multi-call attempt), one parsed cleanly and the other didn't. The first fix version skipped synthesizing a failure signal whenever a real tool call already existed (state.toolCalls.size > 0), on the assumption that meant the model was retrying a LATER, separate attempt after an earlier one already succeeded. That's indistinguishable, from the translator's state, from this same-turn case — so it silently discarded the malformed attempt's information entirely: the client saw the real call succeed and never learned the other tool calls were attempted and rejected. Fix: always synthesize the failure entry when a malformed abort reason is seen, appending it alongside any real tool call rather than skipping it. Multiple tool_calls in one response is normal, well-supported OpenAI behavior (parallel tool calls), so this adds the failure as an additional entry instead of replacing or hiding the real one. Tests (TDD, confirmed red-before-green): - Rewrote the unit test that encoded the old (wrong) assumption to assert both the real and synthesized calls are present. - New fixture (gemini-malformed-function-call-parallel-real-call-stream.json): the real event series from this incident, sanitized. - New integration tests (same file as 3568c7259's) prove both /v1/chat/completions and /v1/responses surface both tool calls correctly from this fixture. Co-authored-by: Markus Hartung <markus.hartung@gmail.com> * feat(sse): honor tool_choice when translating OpenAI requests to Gemini Investigating a live report that gemini-3.1-flash-lite frequently narrates an intended tool call in plain text instead of actually emitting one (dashboard log id 1784591483850-49c408 — 9 raw provider chunks, all plain text, zero functionCall parts, clean finishReason STOP): body.tool_choice was never read anywhere in the OpenAI->Gemini request translator. result.toolConfig was unconditionally hardcoded to { functionCallingConfig: { mode: "VALIDATED" } } whenever tools were present, regardless of what the caller sent. VALIDATED lets the model respond with plain text OR a schema-validated function call at its own discretion — it never forces a call the way OpenAI's tool_choice: "required" (Gemini's ANY mode) does, so a caller had no way to compel a tool call even when explicitly requesting one. Added convertOpenAIToolChoiceToGemini(), mirroring the existing convertOpenAIToolChoice() in openai-to-claude.ts for the same OpenAI tool_choice shapes (string "auto"/"none"/"required", or {type:"function",function:{name}} to force one specific tool): - unset/"auto" -> VALIDATED (unchanged default, no regression) - "required"/"any" -> ANY (forces a call) - "none" -> NONE (disables function calling) - {type:"function",...} -> ANY + allowedFunctionNames: [name] Wired into both Gemini request paths: the direct/base translator (openaiToGeminiBase) and the Antigravity/Cloud Code envelope (wrapInCloudCodeEnvelope), which now reuses the base translator's already- computed toolConfig instead of re-deriving its own hardcoded VALIDATED. This unblocks (but does not itself resolve) the live question — a tool_choice: "required" A/B test against gemini-3.1-flash-lite follows to confirm ANY mode actually changes the narrate-vs-act behavior in practice. Also updates the T11 any-budget allowlist for this file: the "any" string comparisons (tool_choice value "any", not a TypeScript type) are the same documented false-positive pattern already carved out for executors/base.ts. Tests (TDD, confirmed red-before-green): 9 new unit tests covering all tool_choice shapes on both the direct and Antigravity/Cloud Code paths, plus the no-tools and unset-default no-regression cases. Co-authored-by: Markus Hartung <markus.hartung@gmail.com> * chore(quality): file-size baseline for own-growth (#8211) Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: Markus Hartung <markus.hartung@gmail.com> Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Chirag Singhal
|
8565954e65 |
fix(stream): add logging to empty catch blocks in stream error handling (#8143)
* fix(combo,model-fallback,sqljs): three stream-reliability fixes - targetExhaustion: skip remaining same-provider models on 401 auth failure (prevents opencode-zen noauth cascade wasting retry attempts) (#8133) - modelFamilyFallback: skip unsupported models in T5 fallback chain (prevents GitHub provider trying deprecated claude-opus-4.8/4.7) (#8134) - sqljsAdapter: split package.json resolve string to suppress Next.js Can't resolve warning at build time (#8135) * fix(stream): add logging to empty catch blocks in stream error handling - stream.ts: Log errors in onComplete/onFailure callbacks (lines 929, 1112, 2451, 2536, 2561, 2717) - streamHandler.ts: Log errors in stall watchdog and trackPendingRequest (lines 249, 334, 657, 663, 667) - cursor.ts: Add comments to intentional H2 lifecycle catches, log KV/exec errors - next.config.mjs: Externalize sql.js to suppress build warnings Closes #8138, #8139, #8140, #8141, #8142 * refactor(stream): scope PR to logging hygiene, drop out-of-scope exhaustion/fallback hunks Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> * chore(quality): rebaseline stream.ts for #8143 empty-catch logging own-growth Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> Co-authored-by: rafaumeu <rafael.zendron22@gmail.com> Co-authored-by: chirag127 <chirag127@users.noreply.github.com> |
||
|
Austin Liu
|
07dada6b81 |
fix: strip internal reasoning placeholder from user-visible content (#8081) (#8162)
* fix: strip internal reasoning placeholder from user-visible content (#8081) The internal reasoning replay sentinel '(prior reasoning summary unavailable)' can leak into user-visible assistant content when a model echoes it through ordinary message.content / delta.content. Existing suppression only checked reasoning_content fields and reasoning-specific events. Changes: - Add stripInternalReasoningPlaceholder() to reasoningPlaceholder.ts — removes all occurrences of the sentinel and trims; returns '' when nothing meaningful remains - Streaming: strip in responsesTransformer.ts, openai-responses.ts, and openai-to-claude.ts at the delta.content entry point; skip emission entirely when only the placeholder was present - Non-streaming: strip in responseSanitizer.ts sanitizeMessageContent() and sanitizeResponsesMessageContent() (all three text paths) translateText is unaffected (uses mode='translate' via plain newsClient). The per-provider reasoning_content check remains as defense-in-depth. * fix: skip only empty content block on reasoning-placeholder, keep finish_reason/tool_calls (#8081) Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> * chore(quality): rebaseline openai-responses.ts own-growth (#8081 guard) --------- Co-authored-by: Austin Liu <austinliu@Austins-MacBook-Air-3.local> Co-authored-by: Probe Test <probe@example.com> Co-authored-by: Dingding-leo <Dingding-leo@users.noreply.github.com> Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
RCrushMe
|
89bad0fa52 |
fix(responses): close namespace round-trip for Responses-Chat translation (#7936) (#8151)
* fix(responses): close namespace round-trip for Responses-Chat translation (#7936) The #7905 custom-tool-call path landed in release/v3.8.49 but left #7936 open: Responses namespace sub-tools were flattened to a bare leaf on the Chat wire with no response-side closure, so Codex's adjudicator rejected every namespace sub-tool call with `unsupported call` -- it only has a dispatch entry for the header bits (namespace+name), no entry for the bare leaf. This patch closes the round-trip without mutating the Chat wire name (alignment with #7905's bare-leaf contract and with the issue author's proposed fix): * request side (openai-responses.ts): keep tool.function.name as the bare leaf, populate a side-band namespaceToolIdentityMap keyed on that leaf, and thread it through translatedBody._toolNameMap. * request -> response seam (chatCore.ts): extract the identity map before dispatch and pass it through to the non-stream completion path and to all three stream pipelines (translate openai-responses, translate other, passthrough). * response translator (response/openai-responses.ts): in emitToolCall (response.output_item.added) and closeToolCall (custom_tool_call / function_call output_item.done), call resolveRequestToolIdentity() to rewrite the bare leaf back to {namespace,name} and emit codex-compatible independent fields. * passthrough (utils/stream.ts): add a response passthrough rewriter restoreResponsesPassthroughFunctionCallIdentity that intercepts response.output_item.added, response.output_item.done, and response.completed and stamps the same {namespace,name} tuple. * helper (requestToolIdentity.ts): a 20-line stateless resolver; never parses a name. The wire-visible Chat tool.function.name stays the bare leaf -- non-OpenAI providers (NVIDIA, GLM, Kimi, Gemini, ...) frequently truncate or rewrite long __-dotted names; bare leaves avoid that failure mode entirely. The codex ResponseItem::FunctionCall schema (models.rs) declares an independent namespace: Option<String> field and has a function_call_deserializes_optional_namespace round-trip test, so emitting it separately matches the codex adjudicator dispatch. Includes 19 new test cases across 4 files: - request-side bare-leaf wire + side-band ledger construction - response-side tuple emit + unmapped passthrough + apply_patch exclusion - ambiguous-leaf collision safety (entry dropped, leaf emits verbatim) - per-request isolation between concurrent streams - a precompiled Atlassian-style nested namespace override * fix(responses): skip tool_search_call input items instead of 400 (#7936 addendum) Codex 0.42+ emits `tool_search_call` (and later `tool_search_result`) input items when the model uses the dynamic tool-search optimization. They are metadata-only: they record that the model queried a subset of the available tools, and carry nothing that OpenAI Chat Completions can represent. Without an explicit skip in openai-responses.ts, the input loop threw Unsupported Responses API feature: input item type 'tool_search_call' cannot be represented in Chat Completions -- and because these items stay in the Responses API `input` for every follow-up turn, the whole server returned 400 on EVERY subsequent /v1/responses in the same session until the user cleared history. Observed in the wild: /v1/responses 400 "Unsupported Responses API feature: input item type 'tool_search_call' cannot be represented in Chat Completions [longcat/LongCat-2.0 (400), longcat/LongCat-2.0 (400)]" The meituan combo (longcat fallback) was the most visible victim, but the underlying throw is source-format-side and hits any Responses-API consumer whose upstream does not natively support Responses. Fix: stop on the item type the same way `reasoning` is skipped -- display-only metadata, no chat side-effect. Covers both `tool_search_call` and the follow-up `tool_search_result` shapes. Adds 3 unit tests: - tool_search_call is silently skipped (no 400) - tool_search_result is silently skipped - tool_search_call items interspersed with real messages are skipped in order; real messages survive * chore(quality): rebaseline openai-responses.ts + stream.ts own-growth (#7936 namespace round-trip) --------- Co-authored-by: TonPro <hello@tonpro.fu> Co-authored-by: RCrushMe <RCrushMe@users.noreply.github.com> |
||
|
Makcim Ivanov
|
ffb37f99a5 |
fix(cursor): bridge native tools to client calls (#8171)
Co-authored-by: Makcim Ivanov <10184529+makcimbx@users.noreply.github.com> |
||
|
NOXX - Commiter
|
e86e5bcc51 |
feat(media): Adobe Firefly image + video generation provider (#8006)
* feat(media): Adobe Firefly image + video generation provider
Add unofficial adobe-firefly media provider with full OpenAI-compatible
image and video generation: Nano Banana / GPT Image families, Sora 2,
Veo 3.1 (standard/fast/reference), and Kling 3.0.
Supports browser session cookies (auto IMS token exchange) or direct
IMS access tokens, async submit-and-poll against Firefly 3P endpoints,
aspect-ratio and resolution controls, and multi-account web-session UX.
Chat completions are intentionally rejected (media-only surface).
Includes unit coverage for registry wiring, payload builders, auth
resolution, and mocked generate happy-paths.
* fix(adobe-firefly): clio auth, discovery fallback, credits balance
Root-cause 401 invalid token against live firefly.adobe.com captures:
generate/discovery use x-api-key + IMS client_id clio-playground-web
(not projectx_webapp). Align headers/origin, dual cookie to IMS exchange
(clio first, Express fallback), BKS poll rewrite for /jobs/result.
Models: parse POST /v2/models/discovery + static fallback catalog from
adobe/get_models.txt; expand image/video registries.
Limits: GET firefly.adobe.io/v1/credits/balance (SunbreakWebUI1) with
total/remaining + free/plan detail quotas. Clarify cookie vs JWT UX.
Unit tests: 27/27 pass.
* fix(adobe-firefly): reject guest tokens from page-only cookies
Live repro with firefly.adobe.com Cookie export: IMS check with
guest_allowed=true returns account_type=guest (no AdobeID). That token
fails generate (401 invalid token) and credits/balance (403
ErrMismatchOauthToken). guest_allowed=false needs adobelogin.com IMS
session cookies which are not present in a page-only Cookie paste.
- Detect/reject guest JWTs; clear error tells user to paste Bearer JWT
- Prefer user JWT from HAR/mixed paste; improve credential extraction
- Update web-cookie + credential UX to recommend Authorization Bearer
Unit tests 29/29.
* fix(adobe-firefly): production auth, Limits, and 408 load handling
Live validation against firefly.adobe.com + packaged VibeProxy:
Auth / credentials
- Prefer IMS user JWT (Bearer from firefly-3p); reject guest tokens from
page-only cookies with an actionable error
- Extract JWT from Bearer, access_token=, IMS sessionStorage tokenValue,
and mixed HAR pastes; prefer non-guest tokens
- Strip JWT from Cookie header (undici Headers.append crash on mixed paste)
- Keep sherlockToken → x-arp-session-id + sanitized Cookie for generate
Limits
- credits/balance → Record quotas (firefly_total / free / plan) so
providerLimits caches them (arrays were ignored)
- Allowlist adobe-firefly + firefly in USAGE_SUPPORTED + APIKEY limits
- Live: 10000 plan credits parsed end-to-end after refresh
Generate
- Browser-shaped gpt-image body (size auto, no extra top-level size)
- Exponential 408 "system under load" retries (8 attempts) with clear
client message that 408 is Adobe capacity, not invalid token
- Live: generate returns proper 408 under load; balance/models stay 200
Tests: adobe-firefly unit suite 33/33 pass.
* fix(adobe-firefly): match live capture headers; add gpt-image-2
- Do not send firefly.adobe.com Cookie to firefly-3p (wrong-origin; soft 408)
- Lift sherlockToken only into x-arp-session-id
- Poll headers match status_check.txt (Bearer + accept, no x-api-key)
- Catalog gpt-image-2 alias → upstream modelVersion "2" (GPT Image 2)
- Shorter 408 retry budget so clients fail fast with clear message
- Unit suite 34/34
* fix(adobe-firefly): always send x-arp-session-id on generate (fixes 408)
Root cause of Bearer JWT → HTTP 408 colligo "system under load":
submit only set x-arp-session-id when sherlockToken was present in a
cookie paste. JWT-only credentials never sent the header, and Adobe
soft-blocks those requests with instant 408 (x-colligo-timeout:0.0).
A/B against a real user IMS token:
- det nonce + synthetic ARP → 200
- random nonce + synthetic ARP → 200
- det nonce without ARP → 408
Match adobe2api / GPT2Image-Pro:
- buildAdobeSubmitNonce = sha256(user_id + prompt[:256])
- buildAdobeArpSessionId = base64({sid, ftr}) synthetic session
- buildAdobeSubmitHeaders always sets both headers
Live adobeFireflyGenerateImage end-to-end: submit + poll → S3 presigned URL.
* fix(adobe-firefly): drop literal cred fallbacks + type-clean tests
Addresses pre-merge review feedback on #8006:
- Removes the `|| "literal"` fallback after resolvePublicCred() in
adobeFireflyApiKey()/adobeFireflyExpressClientId()/adobeFireflyBalanceApiKey()
(open-sse/services/adobeFireflyClient.ts). resolvePublicCred() already
always returns the decoded embedded default, so the literal fallback
was dead code that reproduced the exact env-or-literal anti-pattern
docs/security/PUBLIC_CREDS.md documents as BAD (Hard Rule #11).
- Replaces the 11 `@typescript-eslint/no-explicit-any` casts in
tests/unit/adobe-firefly.test.ts with concrete types
(Record<string, unknown>, Headers, Error-narrowing on the
assert.rejects predicate), matching the pattern already used
elsewhere in this suite. `no-explicit-any` is a hard ESLint error
under tests/ in this repo.
- Freezes file-size baseline entries for the new
open-sse/services/adobeFireflyClient.ts (1958 LOC, new-file cap 800,
mirrors the qoderCli.ts precedent for a legitimately large new
provider client), open-sse/config/imageRegistry.ts (800->821, new
adobe-firefly registry entry) and the +3 LOC growth in
src/lib/usage/providerLimits.ts (1000->1003).
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
---------
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com>
Co-authored-by: artickc <artickc@users.noreply.github.com>
|
||
|
Paijo
|
3238df3204 |
perf: lazy provider init, P2C quota cache, structuredClone elimination, getSettings→getCachedSettings (batch 2) (#7893)
* perf: startup parallelization, stream TextEncoder lift, auth middleware bottlenecks
Startup (~100-300ms faster cold start):
- Parallelize 4 early imports via Promise.all() in registerNodejs()
- Parallelize 10 independent background services via Promise.allSettled()
- Each service has independent try/catch — no failure domino effect
Streaming pipeline (8 fewer TextEncoder GC allocations per SSE event):
- Lift new TextEncoder() from per-chunk inside buildClaudeStreamingResponse
to function scope alongside existing decoder singleton
Auth middleware bottlenecks (from PerfBottleneckAnalysis):
- Backoff decay loop: replace updateProviderConnection (full CRUD:
SELECT+encrypt+cache-invalidate+backup) with resetConnectionBackoff
(targeted UPDATE of backoff/error columns only)
- Dual .filter() for quota: replace two passes calling
isQuotaExhaustedForRequest per connection with a single for loop
partitioning into withQuota/exhaustedQuota
- Debug-log filter recomputation: capture connectionFilterStatus Map
during the filter pass; debug loop reads 6 string comparisons instead
of 6 function calls per connection
Supporting:
- Add resetConnectionBackoff to src/lib/db/providers.ts (patterned after
clearConnectionErrorIfUnchanged, no CAS check)
- Re-export resetConnectionBackoff from src/lib/localDb.ts
- Update integration-wiring.test.ts regex for parallelized dynamic import
* perf: P2C quota cache, lazy provider init, structuredClone elimination, getSettings→getCachedSettings
- **auth.ts: P2C quota re-evaluation cache** — quotaResults Map threaded
through selectPoolSubset → compareP2CConnections → getP2CConnectionScore.
Populated during filter + partition passes, eliminating redundant
evaluateQuotaLimitPolicy / isQuotaExhaustedForRequest calls when the
P2C comparator re-evaluates previously-scored connections.
- **constants.ts: lazy PROVIDERS via Proxy** — replaces eager
generateLegacyProviders() + loadProviderCredentials() at module load
with Proxy delegating to deferred init on first property access.
- **providerModels.ts: lazy PROVIDER_MODELS + PROVIDER_ID_TO_ALIAS** —
same Proxy pattern for both exports; generateModels()/generateAliasMap()
deferred until first read.
- **stream.ts: structuredClone → minimal object spread** — replaces
O(n) deep clone of SSE response chunks with targeted reconstruction
of only mutated fields (usage, delta.content, finish_reason).
- **progressTracker.ts: TextDecoder lift** — module-level decoder
instead of per-chunk new TextDecoder().
- **Route files: getSettings() → getCachedSettings()** — 13 API route
files converted from uncached per-request DB reads to TTL-cached
wrapper (5s default), eliminating redundant queries on every request.
- **settings.ts: re-export getCachedSettings** from readCache for
non-localDb consumers.
- **Remove settingsCache.ts** — dead file, no imports reference it.
TS compile: 0 errors. Auth tests: 225/225 pass. Services: 269/269 pass.
* perf: Phase 1 tangible wins — egressCache eviction, mmap_size PRAGMA, composite indexes, proxyFallback lazy import
- egressCache: lazy TTL eviction on getCachedEgressIp access (bounds memory
leak to distinct proxy URLs, typically <100)
- mmap_size: apply stored PRAGMA from key_value table (256MiB default) after
applyStoredDatabaseOptimizationSettings — setting was stored but never applied
- schemaColumns: add idx_uh_provider_model_timestamp (covers getModelLatencyStats)
and idx_pc_provider_auth_type (covers 6+ provider_connections queries)
- proxyFallback: convert static import to dynamic import() inside error handler
(defers 210ms module load from startup to first proxy-retry scenario)
* perf: add dedup expression index, unref() sweep timers
- Add COALESCE expression index idx_uh_dedup on usage_history
matching the exact dedup query pattern. Eliminates FULL TABLE
SCAN on every saveRequestUsage insert.
- Add composite idx_uh_provider_model_timestamp on usage_history.
- Add composite idx_pc_provider_auth_type on provider_connections.
- Add .unref() to setInterval in batchProcessor.ts (polling loop).
- Add .unref() to setInterval in runtimeHeartbeat.ts (heartbeat).
* perf: bump SQLite cache_size default from 16MB to 64MB
New installs now start with 64MB page cache (was 16MB). Existing
users' stored settings are unchanged. Reduces disk reads for the
typical ~250MB database by keeping ~25% of pages in memory.
Also resolved pre-existing merge conflict in webhooks.ts.
* docs: add Redis production config guide and proxy port clash investigation report
- docs/redis-production-config.md: comprehensive Redis tuning guide
covering client options, server config, Docker settings, scaling,
and monitoring for all three Redis workloads (rate limiting,
auth cache, quota store)
- docs/proxy-port-clash-report.md: investigation confirming proxy
subsystem has no port binding issues; real EADDRINUSE history
traced to process supervisor crash-loop restart race (#4425) and
live-dashboard port clash (#6324), both already fixed
* fix: address PR #7893 review — add Proxy traps, extract migrations to reduce providers.ts size
- Add set trap to PROVIDER_ID_TO_ALIAS Proxy (providerModels.ts)
- Add deleteProperty traps to all three lazy Proxies (PROVIDERS,
PROVIDER_MODELS, PROVIDER_ID_TO_ALIAS)
- Extract autoMigrateLegacyEncryptedConnections and getGheCopilotHosts
from providers.ts (1129→1036 lines, -93) into providers/migrations.ts
- Both functions re-exported via providers.ts for backward compat
File-size ratchet resolved: src/lib/db/providers.ts now 1036 lines.
* fix: resolve merge conflict markers in 3 route/test files
- model-combo-mappings/route.ts: kept upstream version (Zod pagination
via validateBody + isValidationFailure), restored missing return
statement for GET handler
- playground/presets/route.ts: kept stashed version details (satisfies
type-narrowing + inlined Response) — functionally identical
- error-sanitization.test.ts: matches upstream exactly (no diff)
Test verification: same 7 pre-existing failures confirmed on upstream
baseline (
|
||
|
Jan Leon
|
c1bdd91e7b |
Hide internal reasoning replay placeholders (#7912)
* fix: hide internal reasoning replay placeholder * fix: hide internal reasoning replay placeholder in OpenAI→Claude translation too Mirror the isInternalReasoningPlaceholder() guard already applied to the Responses-API and OpenAI-Responses reasoning paths in openaiToClaudeResponse() (open-sse/translator/response/openai-to-claude.ts). The internal reasoning-replay sentinel was still leaking into the Claude "thinking" content block on this translation path. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> * test: close both-add merge of #7912 and #7905 reasoning/tool tests Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com> Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Jan Leon
|
6b59e814da |
Restore Responses API custom tool calls (#7905)
* fix: restore Responses custom tool calls * fix: preserve nested Responses custom tool calls * fix: reset superseded tool call state * fix: preserve tool precedence and buffered Responses tool arguments * test: verify top-level tool descriptions take precedence * fix: preserve custom Responses tool streaming semantics * test: cover declared custom tool streaming round trips * test: cover Responses custom tool metadata collection * test: cover active Responses custom tool stream * test: isolate Responses active stream regression * fix: complete Responses custom tool round trips |
||
|
backryun
|
00677044be |
[Part 3/3] feat(qwen): add regional Alibaba and Qwen Cloud providers (#7882)
* feat(qwen): add Qwen3.8 Max Preview catalogs [Part 2/3] Rebuilt clean on release/v3.8.49 after Part 1 (#7866) squash-merged — applies only the Part-2 delta (Qwen Web / Qoder qwen3.8-max-preview registration + required-thinking allowlist + Qoder client rework) onto the current tip. No migration in this part (that was Part 1). Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> * feat(qwen): add regional Alibaba and Qwen Cloud providers [Part 3/3] Rebuilt clean on top of Part 2 (#7874) over the current release tip — applies only the Part-3 delta (alibaba Model Studio, Alibaba Token Plan, qwen-cloud, qwen-cloud-token-plan with region selector). No migration in this part. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
backryun
|
65e0aeda79 |
[Part 1/3]refactor(qwen): replace legacy Qwen Code and remove OAuth provider (#7866)
* refactor(cli): remove legacy Qwen Code integration * refactor(qwen): remove deprecated Qwen OAuth provider * feat(cli): rebuild Qwen Code integration for upstream V4 * fix(qwen): clear stale CLI auth on reset * test(qwen): align retired provider coverage * fix(db): renumber qwen-cleanup migration 129 -> 130 release/v3.8.49 tip took slot 129 via #7843 (usage_history_codex_strong_identity, itself renumbered from 128 during the #7838/#7840 base-red cleanup) after this branch forked; renumber remove_unregistered_qwen_data to 130. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: Diego Rodrigues de Sa e Souza <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Ravi Tharuma
|
8b78bc361e |
fix: reserve chat admission before body parsing (#7853)
* fix: reserve chat admission before parsing * fix: release admission on handler failure * fix: reconcile chat admission with the #7862 parse-once route Post-#7862 rebase: the admission-rebuilt request is json()-parsed directly over the already-buffered bytes — the clone()+json() pair is gone, and the parse-once regression tests now pin the post-admission contract (original request: zero json() calls, zero clone() calls; the single materialization is admitChatRequest()'s bounded byte reader). Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: Ravi Tharuma <RaviTharuma@users.noreply.github.com> Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com> Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Andrew B.
|
d8499dacd3 |
fix(stream): emit terminal SSE frames on mid-stream upstream failure (#7699) (#7816)
* fix(stream): emit terminal SSE frames on mid-stream upstream failure (#7699) On /v1/messages (Anthropic format), when the upstream SSE stream fails mid-flight after bytes have been forwarded to the client, OmniRoute used to silently close the connection with no terminal event. Anthropic SDK and Claude Code report "Connection closed mid-response. The response above may be incomplete." Two fixes in open-sse/utils/streamHandler.ts: 1. buildStreamErrorChunks (Claude format) now emits event:message_stop after event:error — the Anthropic stream terminator that clients expect. Previously only event:error was sent, leaving the client hanging. 2. createDisconnectAwareStream pull() now detects upstream "done" without a client-visible terminal marker ([DONE] / response.completed / message_stop) and emits a synthetic terminal error frame instead of silently closing. This covers the case where the upstream drops the connection mid-stream without sending an error chunk. Adds tests/unit/silent-sse-close-7699.test.ts covering both code paths across Claude, OpenAI Chat, and OpenAI Responses formats. Refs: diegosouzapw/OmniRoute#7699 * fix(stream): scope terminal-marker missing detection to known formats with forwarded bytes Gate the done-path synthetic 502 error on bytesWereForwarded AND a known clientResponseFormat. Without this gate, any stream that closes cleanly without a terminal marker (including raw passthrough streams and non-API transforms) is incorrectly treated as a mid-stream drop. - Add bytesWereForwarded flag set on first Uint8Array chunk - Require clientResponseFormat to be set before injecting 502 - Fixes 3 broken stream-handler tests (pipes transformed bytes, slow upstream stall watchdog, normal completion watchdog) - Preserves #7699 fix: Claude-format streams that forwarded content but missed message_stop still get the synthetic terminal frame * fix(stream): scope terminal-marker heuristic to Claude only, add non-SSE regression #7699 is scoped to /v1/messages (Anthropic): Claude clients treat a stream that ends without message_stop as an error, and Anthropic's SSE spec explicitly permits a mid-stream event: error. The issue's own suggested fix says the current OpenAI silent-close "remains reasonable" — so the done-without-terminal-marker synthetic-502 heuristic must not fire for any other clientResponseFormat (gemini/codex/kiro/cursor/openai/openai-responses etc.), where a done stream with no [DONE]/response.completed/message_stop equivalent is not necessarily a silent drop. Narrows the gate from "any truthy clientResponseFormat" to "clientResponseFormat === FORMATS.CLAUDE" specifically. Adds a regression test asserting a plain non-SSE OpenAI-format completion (bytes forwarded, no terminal marker) is NOT mutated with a synthetic error frame. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> * test(stream): trim new regression test to clear the 800-line test-file cap tests/unit/stream-handler.test.ts was 772 lines pre-#7816 (not in the frozen file-size baseline, so it's evaluated as new-file-cap 800). The added regression test pushed it to 807; trim boilerplate to land at 796. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com> |
||
|
Andrew B.
|
2611f5a40b |
fix(stream): synthesize terminal finish_reason chunk when upstream omits it (#7800) (#7804)
* fix(stream): synthesize terminal finish_reason chunk when upstream omits it (#7800) Some providers close the SSE stream without emitting a final chunk carrying a non-null finish_reason. The OpenAI spec requires the terminal chunk to include finish_reason (e.g. "stop"); strict clients (pi CLI) reject the stream with "Stream ended without finish_reason". In passthrough mode (OpenAI Chat Completions shape), track whether a chunk with non-null finish_reason was seen. If not, synthesize a terminal chunk with finish_reason "stop" (or "tool_calls" when tool calls were present) before emitting [DONE]. Translate mode is unaffected — translators (claude-to-openai, gemini- to-openai) already guarantee a terminal finish_reason chunk via finishReasonSent / fallback defaults. Tests: 3 new regression tests covering omission, no-op when upstream sends finish_reason, and tool_calls variant. * fix(stream): track finish_reason in flush path to avoid false synthesis (#7800) The synthetic finish_reason chunk fired even when the upstream DID emit a finish_reason chunk — but as the final buffered line without a trailing newline. The flush path (which handles the leftover buffer) normalized IDs but never set passthroughSawFinishReason, so the synthesis guard !passthroughSawFinishReason stayed true and emitted a spurious synthetic chunk with a chatcmpl-* id, breaking the numeric-id normalization test. |
||
|
Paijo
|
4b06761ad5 |
feat(api): add pagination params to 8 DB modules + recharts code-split (#7046)
* perf: extract recharts into dynamic import wrappers
Bundle recharts behind next/dynamic boundaries to prevent its
large module graph from being included in the initial JS payload.
- CostOverviewTab.tsx → dynamic(() => import('./components/CostCharts'))
- ProviderUtilizationTab.tsx → dynamic(() => import('./components/ProviderCharts'))
- BurnRateChart.tsx → dynamic(() => import('./components/BurnRateChartInner'))
- Created 3 wrapper files with 'use client' and all recharts imports
Reduces initial bundle by ~35 kB (recharts + dependencies).
* perf: add pagination (limit/offset) to apiKeys, combos, providers, provider-nodes
Add optional limit/offset parameters to DB list functions and their
API route handlers. All list functions now return { items, total } when
called with parameters; backward compatible when called without args.
Affected modules:
- lib/db/apiKeys.ts - listApiKeys, getApiKeysByGroup
- lib/db/combos.ts - listCombos
- lib/db/providers.ts - listProviders, getProvidersByGroup
- lib/db/providers/nodes.ts - listProviderNodes, getProviderNodesByGroup
- Corresponding API routes pass through query params
Reduces memory pressure on large datasets by returning one page at a time.
* perf: add pagination (limit/offset) to webhooks, proxies, modelComboMappings, playgroundPresets
Add optional limit/offset parameters to DB list functions and their
API route handlers for the remaining data modules.
Affected modules:
- lib/db/webhooks.ts - getWebhooks returns { webhooks, total }
- lib/db/proxies.ts - listProxies
- lib/db/modelComboMappings.ts - listMappings
- lib/db/playgroundPresets.ts - listPresets
- Corresponding API routes pass through query params
- Re-exports updated: lib/localDb.ts, models/index.ts
Backward compatible: calling without args returns all rows.
* perf: batch pool building and add pagination to quotaPools
Replace per-pool N+1 queries with batch-loading pattern.
- Added batchBuildPools(rows) — collects all pool IDs, does 2 batch
queries (allocations + connections) instead of 2N individual queries
- getPoolsByGroup and listPools now use batchBuildPools
- Added optional limit/offset pagination params
- Fixed SQLite OFFSET-syntax bug: only emit OFFSET when LIMIT also present
- Added quota-pools.test.ts with 10 tests covering pagination edge cases,
batch loading, and the offset-without-limit guard
Reduces pool-page query count from 2N+1 to 3 (constant).
* perf: replace manual offset/limit parsing with Zod paginationSchema in combos GET handler
* fix: replace manual Number()/parseInt pagination with paginationSchema
Endpoints: model-combo-mappings, playground/presets, provider-nodes.
Uses existing Zod schema with z.coerce.number() for proper validation.
* chore: bump proxies.ts frozen baseline 1177->1208 for perf/api-pagination
PR #7046 backward-compatible pagination refactor grew proxies.ts
by +31 lines (1177->1208). Entries return plain array when no
pagination params provided, {items,total} when pagination requested.
* fix(db): finish listProxies()/getWebhooks() pagination shape migration
The pagination refactor changed listProxies(), listPools(),
getModelComboMappings(), listPlaygroundPresets() and getWebhooks() to
return a paginated envelope ({ items, total } / { webhooks, total })
instead of a bare array, but left three real production callers and
several tests on the old array-shaped API:
- src/lib/proxyEgress.ts (validateProxyPool default listProxies impl)
iterated the envelope directly -> "is not iterable" at runtime, hit
by /api/settings/proxies/egress (no injected deps).
- src/lib/proxyHealth/scheduler.ts (sweep()) read proxies.length on the
envelope (undefined), so the health-check sweep silently processed
zero proxies every run.
- open-sse/utils/proxyFallback.ts (getProxyCandidates()) iterated the
envelope inside a try/catch that swallowed the resulting TypeError,
so every user-configured proxy silently vanished from the fallback
candidate list.
Also fixes two TS2558/TS2339 typecheck errors in proxies.ts/webhooks.ts
(db.prepare<T>() generic not supported by this DB wrapper — cast the
query result instead, matching the existing pattern in both files) and
trims one blank re-export separator line in localDb.ts to stay within
the frozen file-size ratchet after 4 new *Count() exports.
Updates the pre-existing unit tests that called the changed functions
directly (db-quota-pools, quota-groups-migration, quota-pool-connections,
quota-pool-delete-prune, db-webhooks, model-combo-mappings-db,
db-playground-presets, db-proxies-crud, proxy-batch-routes-5918,
proxy-registry, error-message-sanitization) to destructure the new
envelope shape instead of treating the result as an array.
Implements the small, well-scoped performance-mark/measure
instrumentation ("omni-pipeline-start"/"omni-pipeline-end"/"omni-pipeline")
that tests/unit/chatcore-streaming-pipeline.test.ts already asserted for
assembleStreamingPipeline() but that had no corresponding source change.
Adds three new regression tests (TDD: each reproduces its bug against
the pre-fix code before the corresponding fix, then passes) covering
the three real production callers above:
tests/unit/proxy-egress-validate-pool-default.test.ts,
tests/unit/proxy-health-scheduler-listproxies-shape.test.ts,
tests/unit/proxy-fallback-candidates-listproxies-shape.test.ts.
Co-authored-by: diegosouzapw <diegosouza.pw@gmail.com>
* fix: resolve rebase conflict in proxies.ts — keep hasBlockingProxyAssignment but drop duplicate extraction leftovers
- Removed duplicate resolveScopePoolInternal, resolveProxyForConnectionFromRegistry,
resolveProxyForScopeFromRegistry already extracted to proxies/rotation.ts
- Removed duplicate hasBlockingProxyAssignment function body already re-exported from proxies/guards.ts
- Removed duplicate PROXY_ALIVE_PREDICATE import
- All typechecks and 45 affected tests pass
* fix(test): account for _reorderConnections in pagination test expectedOrder
createProviderConnection calls _reorderConnections after every insert
which reassigns priorities sequentially. The test was assuming creation
order determines priority order, leading to incorrect expected results.
Fix: query the DB after all inserts and use the actual priority order.
Also removes debug console.log from getRawProviderConnections.
* chore: remove debug tmp-*.mjs files left in PR branch
* test(proxy): migrate the dedup test to the paginated listProxies() shape
#7046 changed listProxies() to return { items, total }, and updated every
production caller plus three of the four test files — tests/unit/proxy-bulk-import-dedup-7594.test.ts
was missed, so its four `listed.length` assertions read `undefined` and the
file went red on the merge train (it passes on the pure release tip).
Test-only: destructure `{ items: listed }` at the four callsites. Verified
proxyEgress.ts needs no change — its local deps shim already unwraps .items,
and tests/unit/proxy-egress-validate-pool-default.test.ts guards exactly that.
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
---------
Co-authored-by: oyi77 <oyi77@users.noreply.github.com>
Co-authored-by: diegosouzapw <diegosouza.pw@gmail.com>
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
|
||
|
Diego Rodrigues de Sa e Souza
|
7b85e1f6f7 | feat(api): sync upstream reasoning.supported_efforts into synced-model catalog (#7694) (#7767) | ||
|
Xiangzhe
|
44e57a1ec1 |
fix(kimi-coding): capture and replay reasoning for thinking-mode turns (#7673)
Kimi Coding (claude-format upstream) never engaged reasoning replay: requiresReasoningReplay() had no kimi-coding/kimi-coding-apikey provider entry and only matched /kimi-k2/i model ids, so thinking was neither captured nor re-injected on multi-turn requests. Additionally, streamed Claude thinking_delta chunks were accumulated into content instead of accumulatedReasoning in createSSEStream, so the reconstructed completion body carried no reasoning_content for the cache to capture. - reasoningCache: add kimi-coding/kimi-coding-apikey providers; broaden model pattern to /kimi[-/]k\d/i (covers k2.6/k2.7 incl. namespaced ids, excludes kimi-latest and non-thinking aliases) - stream: accumulate Claude delta.thinking into accumulatedReasoning so the completion body exposes reasoning_content for replay capture - tests: provider/model predicate cases + a reconstructed-stream-body regression test separating thinking from visible text - docs: sync REASONING_REPLAY provider/pattern lists Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com> |
||
|
Paijo
|
0eed344065 |
perf: Date.now hoist, hasActiveDeltaValue hoist, buffer.split guard in SSE stream (#7066)
* perf: hoist Date.now, hoist hasActiveDeltaValue, avoid per-chunk buffer.split in SSE stream - Hoist to transform entry, remove 2 inner declarations that shadowed the outer one (stream.ts) - Hoist from inline closure to module-level function to avoid allocation per chunk (stream.ts) - Replace unconditional per chunk with -gated split to avoid allocating array when no embedded newline (stream.ts streamHelpers.ts) - Guard to avoid allocating when already at/above limit * fix: correct appendBoundedText slice offset when keep is zero * chore(ci): rebaseline stream.ts 2796->2801 for perf/p1-fixes Add _rebaseline_ entry documenting the +5 line growth from: - |
||
|
NOXX - Commiter
|
7a68a7961c |
fix(notion-web): production-ready labels, multi-workspace, inference, usage (FINAL) (#7768)
* fix(notion-web): use real picker labels as primary model ids Catalog /v1/models now surfaces web-picker names (fable-5, gpt-5.6-sol) instead of Notion food codenames (acai-budino-high, orange-mousse). Food codenames stay internal via notionCodename + resolveNotionCodename for runInferenceTranscript. Legacy codename requests still work; responses echo the client-facing id. Also points discovery/inference at app.notion.com (same host as the AI picker). Follow-up to #7696. * fix(notion-web): explain plan-locked models like Fable 5 Notion returns Fable 5 (acai-budino-high) with isDisabled=true and disabledReason=business_or_enterprise_plan_required. Keep it out of the enabled catalog (requests would fail) but surface a discovery warning so operators know why it is missing. Also warn when space_id is resolved via getSpaces instead of the cookie. * feat(notion-web): auto-detect workspace without pasting space_id Operators only need the raw token_v2 value. When space_id is omitted: - getSpaces loads all workspaces (browser-like headers + user id) - each workspace is probed via getAvailableModels - the richest AI catalog wins Also softens auth hints so they no longer demand a cookie blob with =. * fix(notion-web): pick Business workspace so Fable 5 is listed Probe ALL workspaces instead of early-exiting on the first catalog with >=8 models. Prefer spaces where Fable is enabled over personal spaces where Notion returns isDisabled=business_or_enterprise_plan_required. Cache the chosen spaceId for inference when cookie has no space_id. * fix(notion-web): working inference + honest token estimates - runInferenceTranscript: createThread+threadId, config/context/user transcript, space/user headers (fixes ValidationError 400) - Parse modern NDJSON patch/record-map; strip lang tags - Estimate usage from text (Notion has no metering); mark estimated - Treat all-zero usage as missing; skip USAGE_TOKEN_BUFFER on estimated - Keep estimated flag through response sanitizer (was stripped -> flat 2000) Verified live: fable-5/gpt-5.6-sol chat 200; usage 7 / 65 not constant 2000. * refactor(notion-web): extract helpers to keep discoverNotionWebModels/execute under the complexity cap Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com> Co-authored-by: artickc <artickc@users.noreply.github.com> Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Xiangzhe
|
a9028e9571 |
fix(stream): suppress </think> close marker for Responses API clients (#7747)
* fix(stream): suppress `</think>` close marker for Responses API clients The Claude→OpenAI `</think>` close marker (#4633) exists for Chat Completions clients that scan content for the marker (Claude Code / Cursor). On the openai-responses path the responsesTransformer already maps reasoning_content to structured reasoning items, so the marker has no consumer and leaks verbatim into response.output_text.delta — observed in production with kimi-coding (k3): thinking renders correctly while a stray `</think>` sits at the start of the assistant text (up to 6 consecutive markers when the upstream also emits stray close-tag text deltas). resolveSuppressThinkClose() gains a clientResponseFormat option that always suppresses the marker for openai-responses, winning over both the UA allowlist and an explicit keep header (no legitimate marker consumer exists in the Responses format). chatCore passes the format through, and ExecuteInput now carries clientResponseFormat so the two executors that do their own Claude→OpenAI translation apply the same policy: GLM's Anthropic transport and zed-hosted's Anthropic backend (which previously applied no suppression at all, not even the #5245 UA/header policy). Chat Completions behavior is unchanged (#4633 / #5123 / #5245 / #5312). * refactor(executors): extract helpers to keep execute/executeTransport under the complexity cap Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com> Co-authored-by: xz-dev <xz-dev@users.noreply.github.com> Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Hernan Javier Ardila Sanchez
|
f1a77fefc5 |
fix(combo): auto-clear stale session pins and emit recovery hints on combo exhaustion (#7625)
* chore(ci): add .mergify.yml to main — Mergify only reads config from the default branch (#7168) * fix(ci): add the auto-enqueue pull_request_rule to the Mergify config (queue_conditions alone are eligibility-only) (#7179) * fix(ci): migrate Mergify auto-enqueue to merge_protections_settings.auto_merge_conditions (rules-based path is EOL 2026-07-16) (#7216) * fix(ci): drop Mergify batch settings (batching is a paid-tier feature; free plan queue is serial) (#7220) * fix(ci): merge queue tolerates the advisory dast-smoke failure (its GH-hosted build hang dequeued every attempt) (#7225) * test(ci): make the #6634 selfref guard hermetic — main's copy hard-fails every PR (#7341) main's copy of this test still does git I/O inside a unit test: const baseSrc = git(['show', 'origin/main:' + FILE]); Runners check out a shallow single ref, so origin/main does not resolve and the test dies with 'fatal: invalid object name origin/main'. Every PR into main fails Unit Tests (7/8) on it — today that is #7313, #7315, #7316, #7334, #7336 and #7337, six PRs red on a defect none of them introduced. #7313 has no other red at all. release/v3.8.49 already carries a fix ( |
||
|
Jan Leon
|
74c006e245 |
Add reasoning-based model and effort routing (#7607)
* feat(routing): add reasoning-based model and effort routing * refactor(routing): modularize reasoning and auto-routing pipeline * fix(routing): remove redundant DB re-export and prevent SQL scan false positives * fix(routing): resolve reasoning routing review blockers * fix(i18n): keep release ranking fallbacks outside reasoning * fix(db): renumber reasoning-routing migration past release tip (124→125) 124_generic_session_affinity_ttl.sql (#7274) has since landed on release/v3.8.49 at version 124, colliding with this PR's own 124_reasoning_routing_rules.sql. Renumbers to 125 (the next free slot past the current release tip) and updates the one filename reference in docs/routing/REASONING_ROUTING.md. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> * chore(db): renumber reasoning-routing migration 125→126 (slot taken by #7360) Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> * refactor(api): compact temp-path decls in exportAll GET (complexity-ratchet lines budget) Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> * refactor(api): single-statement auth guard in exportAll GET (function under 80-line cap) Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Hernan Javier Ardila Sanchez
|
9152e3d8f5 |
fix(stream-readiness): bump timeout for heavy Claude-format reasoning replicas (#7612)
* fix(stream-readiness): bump timeout for heavy Claude-format reasoning replicas Third-party Claude-format replicas (Minimax M2.7/M3, ZAI, bailian, agentrouter, wafer, …) inherit Anthropic's stream shape but their reasoning warm-ups routinely exceed the default 80s readiness window — the STREAM_READINESS_TIMEOUT fires before the upstream emits its first non-ping SSE event, surfacing the request as a stalled task to clients like OpenChamber / Claude Code and demanding manual continuation on every long-running task. Mirror the codex_gpt_5_5_high_reasoning +30s bump on every provider whose registry entry has format === 'claude' (excluding first-party claude/anthropic which have stable cold starts). The registry is the single source of truth, so newly-registered replicas inherit the bump without code changes. Stays within the existing maxTimeoutMs cap so a single env knob still bounds the readiness window overall. Tests: - covers Minimax M3, ZAI, official claude/anthropic (no bump), OpenAI (no bump), unknown providers (no false positives), and the maxTimeoutMs cap with the new bump stacked against large payloads. - all 17 stream-readiness-policy tests pass (10 existing + 7 new). - existing 16 stream-readiness + 11 combo-stream-readiness-fallback tests still pass (no regressions). * fix(quality): rebaseline coverage.functions 86.44->86.42 and zizmorFindings 175->176 Pre-existing drift on source branch, not introduced by #7612: - coverage.functions drifted -0.02 from PR #7625 adding failureTracker.ts (+2 function definitions). Coverage denominator grew; numerator unchanged because the 8 coverage shards do not exercise the new file. Legitimate drift from feature addition. - zizmorFindings +1 from upstream workflow drift on release/v3.8.49. PR #7612 touches zero workflow files. Same class as the _rebaseline_2026_07_17_v3849_release rebaseline that bumped 169->175. Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai> --------- Co-authored-by: herjarsa <herjarsa@users.noreply.github.com> Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai> |
||
|
backryun
|
9544fb6353 |
feat(kimi): sync Code, Web, and Moonshot providers (#7531)
* feat(kimi): sync Code, Web, and Moonshot providers * chore(quality): trim frozen file-size overflow in Kimi sync The Kimi/Moonshot provider sync added a net +1 line to both src/sse/services/auth.ts and ProviderDetailPageClient.tsx, pushing each 1 line past its frozen cap in file-size-baseline.json. Drop one optional blank line in each (prettier-neutral, no behavior change) to land back at/under the frozen baseline. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Mrinal Joshi
|
d82ca30253 |
feat(models): advertise Claude reasoning-effort variants in /v1/models (#7497)
* chore(ci): add .mergify.yml to main — Mergify only reads config from the default branch (#7168)
* fix(ci): add the auto-enqueue pull_request_rule to the Mergify config (queue_conditions alone are eligibility-only) (#7179)
* fix(ci): migrate Mergify auto-enqueue to merge_protections_settings.auto_merge_conditions (rules-based path is EOL 2026-07-16) (#7216)
* fix(ci): drop Mergify batch settings (batching is a paid-tier feature; free plan queue is serial) (#7220)
* fix(ci): merge queue tolerates the advisory dast-smoke failure (its GH-hosted build hang dequeued every attempt) (#7225)
* test(ci): make the #6634 selfref guard hermetic — main's copy hard-fails every PR (#7341)
main's copy of this test still does git I/O inside a unit test:
const baseSrc = git(['show', 'origin/main:' + FILE]);
Runners check out a shallow single ref, so origin/main does not resolve and the
test dies with 'fatal: invalid object name origin/main'. Every PR into main
fails Unit Tests (7/8) on it — today that is #7313, #7315, #7316, #7334, #7336
and #7337, six PRs red on a defect none of them introduced. #7313 has no other
red at all.
release/v3.8.49 already carries a fix (
|
||
|
Jan Leon
|
c1a3d83c27 |
fix(combo): reject known context overflow without exhausting providers (#7177)
* Fix context-window exhaustion classification * fix(combo): keep chat.ts/comboStructure.ts under the file-size ratchet + fix context-overflow boundary bug - Extract getKnownContextOverflow (+ its KnownContextOverflow type) out of comboStructure.ts into a new open-sse/services/combo/knownContextOverflow.ts leaf, so the file-size ratchet (cap 800 for new files) passes. - Extract the skipConnectionDisable predicate out of handleSingleModelChat in chat.ts into open-sse/services/combo/comboPredicates.ts::shouldSkipConnDisable, and consolidate the new combo-failure-handling imports, to keep chat.ts under its frozen file-size baseline (1796) after the #7177 request-scoped-failure wiring. - Fix a real boundary bug in getKnownContextOverflow surfaced by the merge: estimateRequestInputTokens counted a caller-omitted `messages: []` (which some combo entrypoints default in) as real content, charging a few phantom "structural" JSON.stringify tokens toward the estimate. That was enough to falsely trip the new known-context-overflow rejection for a request with no real input when max_tokens exactly equals the target's context window (a common config where limit_input === limit_output === limit_context), regressing tests/unit/combo-routing-engine.test.ts's pre-existing #3587 "non-reasoning model does not get max_tokens buffer" case. Empty arrays/objects no longer count as estimable content. - Add a regression test for the exact-boundary empty-content case. Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com> * refactor(combo): move overflow logic into knownContextOverflow module (file-size cap) comboStructure.ts is not frozen in the file-size baseline but is capped at 800 lines; this PR's net +29 on that file alone would push it over once merged. knownContextOverflow.ts already exists in this PR as the dedicated home for "known context limit" logic, so move the genuinely new pieces there instead of leaving them in comboStructure.ts: - hasEstimableContent (new): its own doc comment already frames it purely in terms of the known-context-overflow boundary check, so it belongs next to that check, not in the general request-compatibility file. - getKnownContextLimit (new, requestedOutputTokens-aware): this *is* the "how big is a target's known context window" primitive knownContextOverflow already consumes; hosting it there is a better fit than comboStructure.ts. - getLegacyKnownContextLimit: kept alongside its sibling rather than split across two files, since both are alternate implementations of the same concept (used only by comboStructure.ts's hasKnownCompatibleContextLimit). comboStructure.ts now imports all three back for its own internal callers (estimateRequestInputTokens, getTargetCompatibilityFailures, hasKnownCompatibleContextLimit). deriveRequestCompatibilityRequirements and the RequestCompatibilityRequirements type stay in comboStructure.ts exactly as this PR already has them (still consumed internally there), so knownContextOverflow.ts keeps importing those two, same as before. No behavior change — pure relocation, verified by the existing PR test suite (combo-context-window-filter, combo-breaker-429, combo-failure-log-message, combo-target-exhaustion, diagnostics) plus the pre-existing combo-vision-aware-routing/combo-context-requirements/combo-roundrobin-compat-fallback-6238 suites, all green. Net effect on open-sse/services/combo/comboStructure.ts vs. this PR's merge base: -2 lines (was +29). typecheck:core, lint, and the complexity ratchets (check:complexity, check:cognitive-complexity) are unchanged from this PR's current HEAD — the 4 pre-existing complexity/max-lines findings in valueContainsImagePart/filterTargetsByRequestCompatibility are untouched by this move (same violations, same total ratchet counts, just shifted line numbers). Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com> Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Adam
|
5db2e20c3b |
feat(sse): allow disabling : comment heartbeats via OMNIROUTE_SSE_COMMENTS=off (#7036)
* feat(sse): allow disabling `:` comment heartbeats via OMNIROUTE_SSE_COMMENTS=off * fix(sse): guard process access for edge/Workers + export helper * test(sse): cover sseCommentsEnabled + heartbeat suppression (#7036) |
||
|
Paijo
|
3b7090a1cc |
feat(perf): add performance.mark/measure to SSE pipeline + request-size metric (#7045)
* feat(perf): add performance.mark/measure to SSE pipeline + request-size metric
- streamingPipeline.ts: mark/measure around assembly of SSE transform
chain — 'omni-pipeline-start'/'omni-pipeline-end'/'omni-pipeline'
- stream.ts: compute JSON body byte count on stream creation, emit as
performance.mark('omni-request-body-size', { detail: bytes })
Marks are visible via performance.getEntriesByType('mark') and
performance.getEntriesByType('measure') for DevTools/monitoring.
* fix(perf): prevent memory leak and TextEncoder allocation on hot path
- Add performance.clearMarks/clearMeasures before creating new marks to
prevent timeline accumulation in long-lived processes.
- Replace new TextEncoder().encode(str).length with Buffer.byteLength to
avoid allocating a full Uint8Array just to measure byte length.
* test(perf): add performance instrumentation tests
* chore(ci): rebaseline stream.ts 2796->2805 for perf instrumentation
Add _rebaseline_ entry documenting the +9 line growth from:
-
|
||
|
Diego Rodrigues de Sa e Souza
|
b28331307e |
feat: per-model default reasoning_effort + no-think none on OpenAI path (#6879) (#7631)
Validated in merge-train 2026-07-18 @ 9084b408b: 12198/12201 pass; single red = earlyStreamKeepalive timer test, confirmed load-flake (6/6 green isolated on release tip AND the merged tree; no boarded PR touches keepalive) |
||
|
Diego Rodrigues de Sa e Souza
|
fb612867fd |
feat: add Segmind image+video provider (#6656) (#7608)
* feat(providers): add Segmind image+video provider (#6656) Segmind exposes 200+ hosted image/video models under a single `POST https://api.segmind.com/v1/{model}` REST shape: x-api-key auth, JSON request body, raw media bytes response (no JSON envelope). - New IMAGE_PROVIDERS + VIDEO_PROVIDERS registry entries (format: "segmind") with a curated starter model list (Flux, SDXL, SD3.5, Kandinsky for image; Wan, Hunyuan, LTX, Kling for video). - New connection-metadata entry in specialty-media.ts; segmind added to IMAGE_ONLY_PROVIDER_IDS and VIDEO_PROVIDER_IDS. - Dedicated handlers (imageGeneration/providers/segmind.ts, videoGeneration/providers/segmind.ts) built on a shared REST client (utils/segmindClient.ts) that centralizes the fetch/error/log path so both stay under the complexity/max-lines ratchets. - Extracted the pre-existing Alibaba DashScope video handler out of the frozen videoGeneration.ts into videoGeneration/providers/ dashscope.ts (no behavior change) to make room for the new Segmind dispatch branch under the frozen file-size baseline. - Error responses route through sanitizeErrorMessage() (Hard Rule #12) — verified by dedicated no-leak tests. - Regenerated docs/reference/PROVIDER_REFERENCE.md (250 -> 251 providers) and synced the plain-text provider counts in README.md/ AGENTS.md/CLAUDE.md (anchors/badges left untouched). Tests: tests/unit/segmind-image-video-provider-6656.test.ts (11 cases — registry shape, connection metadata, IMAGE_ONLY/VIDEO_ PROVIDER_IDS membership, mocked-fetch request mapping for both image and video, and sanitized-error-path assertions for both upstream error bodies and network exceptions). No live Segmind key required; response shape (raw media bytes, x-api-key auth) is sourced from https://docs.segmind.com/ and corroborated against https://www.segmind.com/models/flux-schnell/api, https://www.segmind.com/models/sdxl1.0-txt2img/api, and https://www.segmind.com/models/wan2.1-t2v/api. Gates run clean: check-file-size, check:complexity-ratchets (2055/889, both under baseline), typecheck:core, typecheck:noimplicit:core (no new errors), lint (targeted files), check:cycles, check:docs-counts (STRICT provider-count drift resolved), check:docs-sync, check:any-budget:t11, check:tracked-artifacts, check:provider-consistency, check:known-symbols. * test(providers): align APIKEY_PROVIDERS count 167→168 for the new segmind provider (#6656) Adding segmind to specialty-media.ts grows APIKEY_PROVIDERS by one; providers-constants-split.test.ts hardcodes the family-partition total. Legitimate count alignment, not a weakened assertion — all 4 partition/ dedup checks still enforced. |
||
|
Diego Rodrigues de Sa e Souza
|
5bacb719d3 |
feat: add Microsoft Designer as image provider (#6672) (#7609)
* feat(sse): add Microsoft Designer as image provider (#6672) Adds `microsoft-designer-web` — an unofficial, reverse-engineered Bearer-token web-session image provider, modeled on the existing `chatgpt-web`/`copilot-m365-web` "-web" provider category. - Registers the provider in WEB_COOKIE_PROVIDERS (src/shared/constants/ providers/web-cookie.ts) and IMAGE_PROVIDERS (open-sse/config/ imageRegistry.ts, new "designer-web" format). - New handler open-sse/handlers/imageGeneration/providers/designerWeb.ts implements the submit-then-poll DallE.ashx flow (Bearer access_token + ClientId/SessionId/UserId headers -> form POST -> poll for image_urls_thumbnail), wired into handleImageGeneration()'s dispatch. - The upstream ClientId header is a fixed, publicly-shared value (not a secret) — routed through resolvePublicCred() per Hard Rule #11, never as a string literal. - Registers the token-based credential requirement in webSessionCredentials.ts so the provider-connect UI asks for the right field; connection validation falls back to the existing generic web-cookie session-ping validator (no dedicated validator needed). - Extracted the KIE image-model catalog into a co-located open-sse/config/providers/registry/kie/models.ts module (mirrors the existing lmarena/directModels.ts pattern) to keep imageRegistry.ts under the file-size cap while adding the new provider entry. - Regenerated docs/reference/PROVIDER_REFERENCE.md (250 -> 251 providers) and updated the plain-text counts in README.md, AGENTS.md, CLAUDE.md. Tests (tests/unit/microsoft-designer-web-6672.test.ts, 16 cases): registry-entry shape assertions, the resolvePublicCred() shape assertion (Hard Rule #11), and the pure header/form-body/response- parsing helpers plus the handler's submit/poll/error/timeout paths against a mocked fetch — no live Designer session required. Reverse-engineered from the g4f MicrosoftDesigner.py provider reference (researched during #6672 triage); the exact upstream response shape has not been validated against a live Designer session, so the poll-loop implementation follows the documented g4f contract as closely as possible without a live capture. * fix(providers): satisfy web-cookie executor contract + document designer-web env vars (#6672) |
||
|
Diego Rodrigues de Sa e Souza
|
6695cbbf7a |
feat: add EdgeTTS audio-tts provider (#6668) (#7605)
* feat(sse): add EdgeTTS audio-tts provider (#6668) Registers Microsoft Edge "Read Aloud" as a new no-API-key AUDIO_SPEECH_PROVIDERS entry — the first WebSocket-transport TTS provider in the registry. Reverse- engineered/unofficial endpoint, same class of integration already accepted for other "-web" style providers (chatgpt-web.ts, copilot-web.ts). - open-sse/executors/edgeTts.ts: pure Sec-MS-GEC token construction (SHA-256 over a public trusted-client-token + rounded Windows file-time ticks, ported from rany2/edge-tts drm.py), WS message framing (speech.config/ssml), binary-chunk demuxing, SSML building/escaping, and the WS synth call itself (injectable WebSocket ctor for tests, lazy `import("ws")` in production so it never enters esbuild's top-level CJS bundle graph). Per-client-IP sliding-window throttle (SlidingWindowLimiter) since there's no per-user key — one abusive deployment could otherwise get the shared trusted token rate-limited for everyone. - open-sse/utils/publicCreds.ts: embeds the trusted-client-token via resolvePublicCred() (Hard Rule #11) — it's a constant hardcoded in every Edge build and every open-source edge-tts port, not a per-user secret. - Extracted open-sse/utils/audioResponse.ts (shared response helpers) and open-sse/executors/awsPollyTts.ts (AWS Polly handler) out of open-sse/handlers/audioSpeech.ts to stay under its frozen file-size ratchet baseline while making room for the new branch — no behavior change to either extracted piece. - src/app/api/v1/audio/speech/route.ts: thread the caller's IP through to the handler for the new throttle. Tests: tests/unit/edgetts-provider.test.ts (23 cases) — Sec-MS-GEC determinism and cross-check against a hand-derived reference vector, message framing, binary demux, SSML escaping/injection-safety, registry lookup, publicCreds shape, and the error path via an injected fake WebSocket (upstream failure -> sanitized 502, no stack/path leak; Hard Rule #12), plus the per-IP rate limit. No live upstream is required or used — the reverse-engineered protocol can't be validated against real credentials, but every pure/testable seam is covered per the TDD path in the bug/feature validation gate. * test(mutation): register edgetts-provider.test.ts in stryker tap.testFiles (#6668) The new provider's unit test covers a mutated module, so the strict mutation-test-coverage gate requires it in stryker.conf.json's tap.testFiles. Single-line addition (kept the file's existing formatting). |
||
|
Diego Rodrigues de Sa e Souza
|
6f57c88de1 |
fix(sse): silence noisy proxy-failure log on caller-initiated abort (#7266)
* fix(sse): silence noisy proxy-failure log on caller-initiated abort The pinned-proxy dispatch path in proxyFetch.ts logged every failure — including a plain caller abort or the caller's own AbortSignal timeout firing — as "[ProxyFetch] Proxy request failed ... fail-closed". A client cancelling its own request is not a proxy transport failure and shouldn't be misreported as one in ops logs/alerting; it still propagates to the caller unchanged (fail-closed behavior is untouched). Co-authored-by: TuyulSpam <287281626+TuyulSpam@users.noreply.github.com> Inspired-by: https://github.com/decolua/9router/pull/2589 * chore(changelog): fragment for #7266 --------- Co-authored-by: TuyulSpam <287281626+TuyulSpam@users.noreply.github.com> |
||
|
Diego Rodrigues de Sa e Souza
|
6c8392fa45 |
feat(providers): let custom connections opt into prompt-cache capability (#6880) (#7257)
Add a per-connection cache capability override (supportsPromptCaching, cacheControlPassthrough) stored in provider_specific_data.cache, consulted first by providerSupportsCaching() / providerHonorsOpenAIFormatCacheControl() before falling back to the hardcoded CACHING_PROVIDERS name sets. Unblocks prompt_cache_key injection, the compression cache-aware guard, and cache_control passthrough for openai-compatible-chat-<uuid>-style custom connections that can never match the hardcoded provider-name sets. Default (no override) is byte-identical to current behavior. |
||
|
Diego Rodrigues de Sa e Souza
|
b9433fd03a |
fix(sse): reconstruct Claude-format content in synthetic bypass responses (#7248)
* fix(sse): reconstruct Claude-format content in synthetic bypass responses
handleBypassRequest() returns a canned response for CLI warmup/title-
extraction patterns without calling the provider. For Claude-format
clients (e.g. Claude Code CLI), the non-streaming path merged translated
SSE chunks by taking message_start.message as-is — but the
openai-to-claude translator always initializes that message with
content: [] and streams the actual text via separate
content_block_start/delta events. Every synthetic Claude-format
bypass response therefore silently returned empty content.
mergeChunksToResponse() now rebuilds the content array from
content_block_start/delta events (mirroring the streaming path) and
carries over stop_reason/stop_sequence from message_delta. Extracted
the response-builder helpers (createOpenAIResponse,
create{Non}StreamingResponse, mergeChunksToResponse) out of
bypassHandler.ts into a new open-sse/utils/bypassResponse.ts module so
this logic has a single owner instead of being duplicated inline.
Co-authored-by: KunN-21 <kunn21.nv@gmail.com>
Inspired-by: https://github.com/decolua/9router/pull/2404
* chore(changelog): fragment for #7248
* refactor(sse): extract Claude chunk-merge helpers to fix complexity ratchet
mergeChunksToResponse() regressed both quality ratchets by +1
(complexity 2057>2056, cognitive 891>890). Split the Claude-format
reconstruction into buildClaudeContentBlocks(), applyClaudeMessageDelta()
and mergeClaudeChunks() — same behavior, verified by the existing
bypass-response-claude-merge.test.ts (4/4 passing unchanged).
---------
Co-authored-by: KunN-21 <kunn21.nv@gmail.com>
|
||
|
Diego Rodrigues de Sa e Souza
|
ef777d77e1 |
feat: editable ComfyUI base-URL field + per-connection override for image/video/music generation (#6928) (#7232)
* feat(providers): editable ComfyUI base-URL field + per-connection override for image/video/music generation (#6928) Adds a shared resolveComfyUiBaseUrl() helper (open-sse/utils/comfyuiClient.ts) that prefers a per-connection providerSpecificData.baseUrl override over the registry default, wired through the comfyui dispatch branch in the image, video, and music generation handlers plus a best-effort authType:"none" credential lookup in all three /v1/{images,videos,music}/generations routes (never hard-fails when no connection exists, so zero-config localhost users are unaffected). Surfaces an editable base-URL field on the ComfyUI connection form by adding it to CONFIGURABLE_BASE_URL_PROVIDERS / DEFAULT_PROVIDER_BASE_URLS / getProviderBaseUrlPlaceholder in providerPageHelpers.ts, so Docker-network setups (e.g. http://comfyui:8188) can be configured the same way self-hosted chat providers are. Closes #6928 * refactor(6928): extract local-override credential lookup in media routes The inline per-connection override block nested if>if inside the music and videos POST handlers, taking each to cognitive complexity 16 (>15) — two NEW violations that broke check:complexity-ratchets (892 > baseline 890). Extracted resolveLocalOverrideCredentials() in both routes; behaviour is unchanged. cognitive-complexity back to 890 = baseline. |
||
|
Diego Rodrigues de Sa e Souza
|
6b0c295b95 |
fix(sse): stop per-byte enumeration of binary image bytes in log redaction (#7297) (#7576)
captureCurrentProviderRequest mirrors every Bedrock Converse request into the pending-request log tracker right after openAIToBedrockConverse() builds it, including the decoded image.source.bytes Uint8Array. sanitizePayloadPII() and redactPayload() in src/lib/logPayloads.ts both gate their recursive walk on Array.isArray(), which is false for typed arrays, so each image fell into the generic-object branch and got enumerated one JS key per decoded byte (twice, once per function) before any truncation bound applied. For 3x ~1MB images this took ~4s of synchronous, event-loop-blocking work, matching the reporter's "1-2 images OK, 3+ fails" threshold and their --stack-size observation (data-width pressure, not call-depth). Add an opaque-binary short-circuit (ArrayBuffer.isView) ahead of the Array.isArray branch in both functions, returning a fixed-size placeholder instead of recursing. Apply the same guard to cloneBoundedForLog() in open-sse/utils/requestLogger.ts for defense-in-depth (same blind spot, only accidentally safe today via its own key-count slice). Regression test reproduces the exact reporter shape (3x 1MB images) through the real openAIToBedrockConverse() converter and protectPayloadForLog(), asserting completion well under the previous ~4s and that binary bytes are never expanded into per-byte object keys. |
||
|
Diego Rodrigues de Sa e Souza
|
4de52c6e7c |
fix(sse): split effort/reasoning suffix off pinned cursor model ids (#7289) (#7577)
resolveRequestedModel() only special-cased "auto" and the composer
"-fast" suffix; every pinned Claude/GPT id carrying an effort/reasoning
suffix (e.g. "claude-opus-4-8-high", "gpt-5.5-high") fell through and
was sent to cursor's server verbatim as model_id, with an empty
parameters array. Cursor has no route for the suffixed id -- it only
knows the base id plus an out-of-band ModelParameter -- so it accepted
the request but returned an empty turn.
Split the known effort suffixes (-low/-medium/-high/-xhigh/-max) off
the base id: Claude ids surface an {id:"effort", value} parameter,
GPT ids surface {id:"reasoning", value}, matching the real cursor-agent
client's wire format. encodeAgentRunRequest()'s ModelDetails fields
derive from the same resolved base id, so the #3714 pinned-model
ModelDetails envelope stays correct without further changes.
Updates the existing resolveRequestedModel test that locked in the
buggy verbatim pass-through, and the #3714 ModelDetails test to assert
against the base id. Adds a dedicated regression test file proving the
Claude/GPT split plus non-regression of the "-fast" toggle and
unsuffixed ids.
|
||
|
Diego Rodrigues de Sa e Souza
|
a1299d2aba |
fix(sse): combo failover for OpenAI streams truncated without finish_reason (#7285) (#7568)
validateResponseQuality() only recognized Claude SSE lifecycle events (message_start/content_block_*/message_stop/message_delta.stop_reason). An OpenAI-shape stream (choices[].delta) that emits some bytes (e.g. a role-only delta) and then closes without ever carrying finish_reason (and without a data: [DONE] sentinel) fell through to the generic replay branch and was forwarded to the client as a success instead of triggering combo failover. Adds OpenAI-shape lifecycle tracking (hasChoicePayload/hasTerminalMarker) parallel to the existing Claude tracking: when an OpenAI-shape chunk was seen but the stream ends without finish_reason or [DONE], and no recognized content was found, mark the response invalid so combo failover retries a sibling target. Healthy OpenAI streams (finish_reason present, or real content found) are unaffected — they exit the peek loop before reaching this check, preserving the #3399/#3685 pass-through contract. Regression test: tests/unit/combo-streaming-openai-no-finish-reason-7285.test.ts |
||
|
KooshaPari
|
fd468b5ef1 |
Use OpenAI chunks for early chat keepalives (#7136)
* Use OpenAI chunks for early chat keepalives * Update keepalive assertion to match chat completion chunk format --------- Co-authored-by: Diego Rodrigues de Sa e Souza <8016841+diegosouzapw@users.noreply.github.com> |
||
|
Diego Rodrigues de Sa e Souza
|
0130a4bbb2 |
fix(sse): handle space-separated arg name/value in Composer tool calls (port from 9router#1811) (#7116)
parseInnerCall only split arg segments on a newline between the arg name and its value. Cursor's live Composer/Auto output has been observed using a single space instead, so those segments were treated as one long (space-containing) arg name with an empty value, silently no-opping Write/tool calls for Composer/Auto models. Fall back to splitting on the first whitespace boundary when no newline is present in the segment. Reported-by: way-art (https://github.com/decolua/9router/issues/1811) |
||
|
Diego Rodrigues de Sa e Souza
|
60448d4f31 |
fix(executors): forward X-Session-ID/X-Title agent metadata headers (#7104)
* fix(executors): forward X-Session-ID/X-Title agent metadata headers (port from 9router#2413) Custom agent clients (e.g. non-OpenCode providers) commonly send X-Session-ID and X-Title headers for upstream request tracking/attribution, but forwardOpencodeClientHeaders() only forwarded x-opencode-* keys plus User-Agent, silently dropping these for every client. Extends the existing case-insensitive allowlist forwarding path with x-session-id/x-title. Reported-by: Atikur Rahman Chitholian (@chitholian) (https://github.com/decolua/9router/issues/2413) * chore(changelog): move #2413 entry to changelog.d fragment Consistency with the repo's canonical changelog.d/fixes/ workflow (avoids merge-storm re-conflicts from editing CHANGELOG.md directly). |
||
|
Diego Rodrigues de Sa e Souza
|
f8ef562658 |
fix(sse): recognize xiaomi-tokenplan mimo as a thinking-mode model (#7098)
* fix(sse): recognize xiaomi-tokenplan mimo as a thinking-mode model (port from 9router#1321) The reasoning_content injector already handles DeepSeek/Kimi/K2/MiniMax thinking-mode upstreams, echoing a placeholder reasoning_content on assistant turns that lack one. Its THINKING_MODEL_PATTERNS list omitted the xiaomi-tokenplan mimo family, so requests through xiaomi-tokenplan/mimo-v2.5-pro still hit upstream's 400 'reasoning_content in the thinking mode must be passed back to the API', making the model unusable in multi-turn conversations (e.g. Codex CLI). Add a /\bmimo\b/i pattern so mimo models get the same treatment. Reported-by: z.wl (@xxue-z) (https://github.com/decolua/9router/issues/1321) * docs(changelog): add fragment for #7098 mimo thinking-model fix |
||
|
Diego Rodrigues de Sa e Souza
|
07e1011d3b |
fix(stream): reconcile encrypted Codex reasoning visibility without mutating upstream item (#7304)
* fix(stream): reconcile encrypted Codex reasoning visibility without mutating upstream item Resolves the collision between two open PRs on ensureVisibleResponsesReasoningSummary: #7095 (xz-dev) found that chat clients see nothing when Codex exposes reasoning only as encrypted_content, and added a visible placeholder — but did so by mutating item.summary in place. #7176 (JxnLexn) found that same mutation corrupts the forwarded response item, discarding the encrypted_content shape Codex needs for follow-up requests, and removed the mutation — but that also silently dropped the placeholder, so chat clients went back to seeing nothing. The mutation existed only so a later line could read the summary text back off the same item. getVisibleResponsesReasoningSummaryText() computes that text without touching the item, so: - synthetic response.reasoning_summary_text.delta / .part.done events still carry the placeholder for chat clients (#7095's goal), and - the forwarded response.output_item.done payload keeps its original encrypted_content intact with no fabricated summary field (#7176's goal). Applied at both call sites #7095 identified: the native Responses passthrough in stream.ts/passthroughTailProcessor.ts, and the Responses-to-Chat-Completions translator in openai-responses.ts. Closes #7095, closes #7176. Co-authored-by: Xiangzhe <xiangzhedev@gmail.com> Co-authored-by: Jan Leon <Jan.gaschler@gmail.com> * test(stream): guard the encrypted-reasoning mutation via the completed backfill path The output_item.done line is echoed verbatim on the wire, so a re-introduced item.summary mutation does NOT surface in that event — verified by re-injecting the mutation, which left the existing assertion green. The mutation does surface in the response.completed snapshot, where the captured reasoning item is re-serialized when upstream sends an empty output (store: false). Adds that case, which fails as expected when the mutation is re-introduced, making the #7176 half of the reconciliation an enforced regression guard rather than an incidental property of the current code path. Co-authored-by: Xiangzhe <xiangzhedev@gmail.com> Co-authored-by: Jan Leon <Jan.gaschler@gmail.com> --------- Co-authored-by: Xiangzhe <xiangzhedev@gmail.com> Co-authored-by: Jan Leon <Jan.gaschler@gmail.com> |
||
|
Diego Rodrigues de Sa e Souza
|
de2b464c3f | fix: sanitize non-Latin1 chars in combo diagnostic headers (#6612) (#7190) | ||
|
Diego Rodrigues de Sa e Souza
|
e078664ddd |
fix(sse): schema-aware optional tool-arg normalization for Codex routes (#6951) (#6992)
stripEmptyOptionalToolArgs was allowlist-only (Read/Subagent) and only stripped empty-string/empty-array values, so Responses API strict mode (every property forced into `required`) could forward a forced non-empty value (e.g. Agent.isolation) or a schema-declared default value verbatim to the client. Add schema-aware drop-if-default and generalized drop-if-empty (any tool, gated on schema.required), and thread each tool's JSON Schema from the request's tools[] into the two streaming call sites (response.output_item.done handling). Closes #6951 |
||
|
Diego Rodrigues de Sa e Souza
|
94328d5cb2 |
fix(sse): apply commentary-phase drop filter in TRANSLATE mode (#6952) (#6990)
The #6199/#6561 commentary-phase filter (shouldDropResponsesCommentaryEvent) was wired only into createSSEStream's PASSTHROUGH branch. The TRANSLATE-mode loop (openai-responses upstream -> another client format, e.g. codex routes streaming into Claude Code) called translateResponse() on every raw chunk without checking phase, so internal commentary-phase scratchpad text leaked into the client-visible content channel as duplicate prose and narrated tool-call arguments. Extends the same stateful filter into TRANSLATE mode via a small factory (createTranslateCommentaryFilter) that owns its own item/index Sets, keeping the wiring in stream.ts (a frozen file) to a single guarded line. Fail->pass evidence: - tests/unit/repro-6952-commentary.test.ts against origin/HEAD (pre-fix): FAILED - "commentary-phase prose must not reach the translated client stream" - Same test against the fix: PASSED (2/2) |