fix(antigravity): strip Anthropic-only output_config from Cloud Code envelope (port from 9router#1944) (#4431)

Integrated into release/v3.8.32

CHANGELOG.md

@@ -24,6 +24,7 @@ _In development — bullets added per PR; finalized at release._
 - **fix(translator):** strip Claude output_config before MiniMax (rejected upstream). (thanks @hiepau1231)
 - **fix(translator): OpenAI audio input now reaches Gemini/Antigravity instead of being silently dropped** — `input_audio`/`audio` content parts on the OpenAI→Gemini path matched no handler in `convertOpenAIContentToParts` and were discarded with no error. They are now mapped to a Gemini `inlineData` part with an `audio/<format>` mime type (wav, mp3, …). (thanks @mugnimaestra)
 - **fix(combo): model lockout now honors a long upstream quota reset instead of retrying within minutes** — when a combo target returned a quota error carrying an explicit long reset (e.g. Antigravity `Resets in 160h27m24s`, a `Retry-After` header), the per-model lockout capped at the short base cooldown (~minutes) and discarded the parsed reset, so the exhausted model kept being retried far too early. The lockout now applies the parsed reset when it exceeds the base cooldown, and the Antigravity error-message parser also matches the plural `Resets in …` phrasing. (thanks @Ansh7473)
+- **fix(antigravity): Claude models no longer 400 with `Unknown name "output_config"`** — Anthropic/Claude-Code-only fields (`output_config`, legacy `output_format`) leaked into the Google Cloud Code request envelope via its top-level field passthrough, and Google rejects unknown envelope fields with `400 Invalid JSON payload received. Unknown name "output_config"` — breaking every Claude model served through Antigravity in IDEs. Those fields are now dropped before the envelope is built. (thanks @Duongkhanhtool)
 - **fix(combo): round-robin members fail over faster under concurrency saturation via a configurable queue depth** — when a round-robin combo member was saturated, requests sat in the per-model semaphore's **unbounded** queue and only failed over to the next member after the full `queueTimeoutMs` (default 30s) elapsed — so a burst of agentic requests deep-queued one hot member instead of spilling to healthy ones. The per-model semaphore now accepts a bounded queue depth and emits `SEMAPHORE_QUEUE_FULL` once it is full (the round-robin loop already cascades on that code), so a configured low depth fails over immediately. A new `queueDepth` combo-config knob (global default / provider override / per-combo, default **20** for backward compatibility; **0** = never queue → fail over now) is exposed in Settings → Combo Defaults. ([#3872](https://github.com/diegosouzapw/OmniRoute/issues/3872) — thanks @KooshaPari)
 - **fix(pricing): align Claude Code (`cc`) pricing with current Anthropic per-MTok rates** — the `cc` provider block in the default pricing table had stale numbers across every Claude 4.x family entry — most visibly, `claude-opus-4-5-20251101` was billed at the deprecated Opus 4.1 rate (`input $15` / `output $75`), and `claude-haiku-4-5-20251001` was at half the current Haiku 4.5 rate. The `cached` (cache hit) and `cache_creation` (5-minute cache write) multipliers were also off across Opus 4.6/4.7/4.8, Sonnet 4.5/4.6, Haiku 4.5, and Fable 5. All eight entries now match the rates Anthropic publishes (input, 5m cache write at 1.25x input, cache hit at 0.1x input, output; reasoning billed at the output rate), so cost accounting on the dashboard and per-request usage events stop under- or over-reporting Claude Code spend. (thanks @chulanpro5)
 - **fix(executors): sanitize Anthropic-shape content parts before GitHub Copilot `/chat/completions`** — Claude models on GitHub Copilot driven from clients like Cursor IDE (e.g. `gh/claude-sonnet-4.6`) failed with `Provider returned error: type has to be either 'image_url' or 'text' (reset after 30s)` because the client passed through Anthropic-shape content parts (`tool_use`, `tool_result`, `thinking`) untouched, and the Copilot chat-completions endpoint only accepts `text`/`image_url`. `GithubExecutor.transformRequest` now serializes any unsupported part type as `text` (preserving the model's context), drops empty parts, and collapses to `null` when an assistant message's only content was tool_calls — `tool_calls` ride alongside untouched. Codex-family models still route through `/responses` unchanged. (thanks @cngznNN)

config/quality/file-size-baseline.json

@@ -2,6 +2,7 @@
   "_comment": "Catraca de tamanho (check-file-size.mjs). frozen so pode encolher; arquivos novos <= cap. --update ratcheta.",
   "_rebaseline_2026_06_20_4023_web_cookie_noauth_validation": "PR #4023 (oyi77) own growth: src/lib/providers/validation.ts 4450->4518 (+68 = a new validateWebCookieProvider that probes the provider's /models endpoint — 401/403 => AUTH_007 SESSION_EXPIRED, any other status => valid session, empty cookie => invalid, provider-not-in-registry => unsupported — plus a local STANDARD_USER_AGENT const for the probe). Cohesive validator at the validateProviderApiKey dispatch; not extractable. Covered by tests/unit/provider-validation-web-cookie-auth007.test.ts. Heavily curated on merge — the PR's branch was badly stale-based (squash-base-stale), so its tree was DESTRUCTIVE: providers/index.ts deleted live providers openadapter/dit/tokenrouter (added by #4313) and the executor/base.ts edits reverted release fixes (#4037 duckduckgo host, theoldllm gpt5 models, base.ts fetch-start-timeout). Only the purely-additive validation feature was kept (validation.ts validateWebCookieProvider + errorCodes AUTH_007 + the test). Dropped: 5 malformed new registry entries (used non-RegistryEntry fields defaultModel/auth + referenced non-existent executors -> tsc TS2353), the destructive providers/index.ts + executor reverts, the unrelated pr-*.sh automation scripts, and evals/types.ts (belongs to the deferred evals modularization #4422). Also removed the PR's fragile 'Phase 2' executor probe (ran a live upstream chat during validation + classified any 'auth'-containing error as SESSION_EXPIRED) and rewrote the test to install its fetch mock before module load (the original mocked too late and silently hit live chatgpt.com).",
   "_rebaseline_2026_06_20_1308_model_lockout_honors_reset": "port from 9router#1308 own growth: open-sse/services/accountFallback.ts 1731->1752 (+21 = the new exported pure helper selectLockoutCooldownMs + its doc comment — picks the parsed upstream reset as the model-lockout exactCooldownMs when it exceeds the base cooldown, e.g. Antigravity \"Resets in 160h\", else preserves the existing 0/base behavior) and open-sse/executors/antigravity.ts 1680->1686 (this PR +1 = parseRetryFromErrorMessage regex `reset` -> `resets?` so plural \"Resets in 160h27m24s\" matches, plus a comment line; frozen set to the SUM 1686 with the concurrent #1944 which adds +5 at the disjoint passthroughFields region of the same file, so either merge order passes — pair-file rule). The combo lockout call sites in combo.ts now pass selectLockoutCooldownMs(cooldownMs, mlSettings) instead of always base/exponential, so an exhausted model honors the real upstream reset instead of being retried within minutes. Both edits are cohesive at the existing lockout/parse chokepoints; the helper is its own pure function (not extractable further). Covered by tests/unit/combo-model-lockout-honors-reset-1308.test.ts.",
+  "_rebaseline_2026_06_20_1944_antigravity_strip_output_config": "port from 9router#1944: open-sse/executors/antigravity.ts frozen set to the measured cumulative 1687 of two concurrent PRs that touch disjoint regions of this file, so either merge order passes (pair-file rule). #1944 adds +6 at the envelope passthroughFields destructuring (~line 759: drop output_config/output_format — Anthropic/Claude-Code-only fields that Google's Cloud Code envelope rejects with `400 Unknown name \"output_config\"`, which broke every Claude model on Antigravity); #1308 adds +1 at parseRetryFromErrorMessage (~line 889: regex reset->resets?). Base 1680 + 6 + 1 = 1687 (re-measured on the real merge tip — the earlier 1686 estimate was off by one). Both edits are cohesive at their chokepoints; not extractable. Covered by tests/unit/antigravity-strip-output-config-1944.test.ts.",
   "_rebaseline_2026_06_20_3368_pool_tools": "PR #3368 own growth: open-sse/mcp-server/server.ts 1468->1509 (+41 = register the web-session pool observability tools — poolStatus/poolSessions/poolWarm — at the existing createMcpServer registration chokepoint, with their read:health/write:resilience scope wiring). Reconciled late: the server.ts bump was a separate commit that did not ride along when the feature commit was cherry-picked onto release/v3.8.32 (squash-base-stale rebuild). Thin cohesive registration at the single tool-registration boundary; not extractable. Covered by tests/unit/mcp-pool-tools-3368.test.ts.",
   "_rebaseline_2026_06_20_4401_webfetch_validators": "PR #4401 own growth: src/lib/providers/validation.ts 4428->4450 (+22 = two new API-key validators — Firecrawl + Jina Reader — each a cohesive provider validator branch mirroring the existing ones; 401/403->invalid else->valid). Not extractable (it IS the per-provider validator list). Covered by tests/unit/provider-validation-webfetch-4401.test.ts.",
   "_rebaseline_2026_06_20_4380_parse_once": "PR #4380 own growth: src/sse/handlers/chat.ts 1486->1491 (+5 = thread the once-parsed request body from the route guard into handleChat, replacing the duplicate re-parse). The reusable body accessor lives in the new src/sse/handlers/requestBody.ts (<cap). Thin wiring at the single entry chokepoint; not extractable. Covered by tests/unit/chat-request-body-parse-once-4380.test.ts.",
@@ -88,7 +89,7 @@
   "frozen": {
     "open-sse/translator/request/openai-to-kiro.ts": 807,
     "open-sse/config/providerRegistry.ts": 4731,
-    "open-sse/executors/antigravity.ts": 1686,
+    "open-sse/executors/antigravity.ts": 1687,
     "open-sse/executors/base.ts": 1399,
     "open-sse/executors/chatgpt-web.ts": 2870,
     "open-sse/executors/claude-web.ts": 1057,

open-sse/executors/antigravity.ts

@@ -763,6 +763,12 @@ export class AntigravityExecutor extends BaseExecutor {
       requestType: _requestType,
       requestId: _requestId,
       request: _request,
+      // #1944: output_config (and the legacy output_format) are Anthropic/Claude-Code-only
+      // fields. Google's Cloud Code envelope rejects unknown top-level fields with a 400
+      // ("Invalid JSON payload received. Unknown name \"output_config\""), which broke every
+      // Claude model served via Antigravity. Drop them so they never reach the envelope.
+      output_config: _outputConfig,
+      output_format: _outputFormat,
       ...passthroughFields
     } = normalizedBody;
 

tests/unit/antigravity-strip-output-config-1944.test.ts

@@ -0,0 +1,45 @@
+import { test } from "node:test";
+import assert from "node:assert/strict";
+
+import { AntigravityExecutor } from "../../open-sse/executors/antigravity.ts";
+
+// Regression for #1944: Claude models served via Antigravity (Google Cloud Code) started
+// returning `400 Invalid JSON payload received. Unknown name "output_config"` (path "/").
+// `output_config` is an Anthropic/Claude-Code-only field; Google's Cloud Code envelope
+// rejects unknown top-level fields. The Antigravity envelope spreads `...passthroughFields`
+// (every top-level body field except a known few), so a top-level `output_config` (or the
+// legacy `output_format`) leaked straight into the envelope Google validates.
+
+test("transformRequest drops a top-level output_config from the Antigravity envelope (#1944)", async () => {
+  const executor = new AntigravityExecutor();
+  const body = {
+    output_config: { effort: "high", format: { type: "json_schema" } },
+    output_format: "json",
+    request: {
+      contents: [{ role: "user", parts: [{ text: "hi" }] }],
+      generationConfig: {},
+    },
+  };
+
+  const result = await executor.transformRequest("antigravity/claude-sonnet-4-6", body, true, {
+    projectId: "project-1",
+  });
+
+  if (result instanceof Response) throw new Error("Unexpected Response from transformRequest");
+  assert.equal(
+    (result as Record<string, unknown>).output_config,
+    undefined,
+    "output_config must not reach the Google Cloud Code envelope"
+  );
+  assert.equal(
+    (result as Record<string, unknown>).output_format,
+    undefined,
+    "legacy output_format must not reach the envelope either"
+  );
+  // The sanitized inner request (Claude path) must also be free of output_config.
+  assert.equal(
+    (result.request as Record<string, unknown>).output_config,
+    undefined,
+    "output_config must not survive in the inner Gemini request"
+  );
+});