Repository navigation
Closed
Description
Before submitting
- I searched existing issues and did not find a duplicate. (Related: fix(server): Antigravity 1.5 quota errors resume at the reset #39196, fix(server): Antigravity resumes a turn after an empty model reply #39123, fix(server): Antigravity 502s fail the turn with the real error #33596, [Bug]: Antigravity provider exposes Unavailable models (Neon/Orchestrator), causing false quota crashes #39438; upstream: [Bug]: shouldReplaceUserErrorMessage & (*ACPServer).sendTurnStepUpdates mask 4xx/502 errors as generic '⚠ Agent execution terminated due to error.' and skip retry on transient 404 google-antigravity/antigravity-cli#1485)
- I included enough detail to reproduce or investigate the problem.
Area
apps/server
Steps to reproduce
- Start a multi-step Antigravity thread on
Gemini 4 Neon (High)(MODEL_PLACEHOLDER_M532) using Antigravity ACP runtime1.5.13. - Run a long agentic turn with multiple tool calls.
- Intermittently (mid-turn, after 10+ successful
streamGenerateContentcalls on the same session), Google'sdaily-cloudcode-paESF edge returns a transientHTTP 404 Not Found(NOT_FOUND: Requested entity was not found.).
Expected behavior
AntigravityTurnOutcome.tsshould parse Antigravity 1.5'sSTATUS (code NNN): messageerror step shape (similar to fix(server): Antigravity 1.5 quota errors resume at the reset #39196) instead of falling through to the generic"Antigravity stopped this turn because of an error."message.- When a
404 NOT_FOUNDhappens mid-turn (i.e. after prior steps in the session already succeeded on this model, distinguishing it from permanent unentitled/invalid-model 404s like [Bug]: Antigravity provider exposes Unavailable models (Neon/Orchestrator), causing false quota crashes #39438), T3 Code should ideally treat it as retryable / perform a bounded auto-resume (similar to fix(server): Antigravity resumes a turn after an empty model reply #39123) rather than aborting the turn — sinceagyitself does not retry HTTP 404 ([Bug]: shouldReplaceUserErrorMessage & (*ACPServer).sendTurnStepUpdates mask 4xx/502 errors as generic '⚠ Agent execution terminated due to error.' and skip retry on transient 404 google-antigravity/antigravity-cli#1485).
Actual behavior
The assistant message emits:
⚠ NOT_FOUND (code 404): Requested entity was not found.
and T3 Code terminates the turn with:
Provider error
Antigravity stopped this turn because of an error.
In apps/server/src/provider/acp/AntigravityTurnOutcome.ts:
API_ERROR(/Error (\d{3}), Message: (.+?), Status: ([A-Z_]+)/s) does not match Antigravity 1.5'sNOT_FOUND (code 404): Requested entity was not found.payload, soapiErrorisnulland it falls through to the generic non-retryableprovider_error.- Even with fix(server): Antigravity 1.5 quota errors resume at the reset #39196's
STATUS_ERRORparser,retryable: rateLimited || Number(httpStatus) >= 500would still leave transient mid-turn404s non-retryable.
Sending continue immediately resumes the exact same session and succeeds for dozens more steps.
Impact
Major degradation or frequent failure
Version or commit
0.1.12-nightly.20261120.4994 (Antigravity ACP runtime 1.5.13)
Environment
macOS 26.3 (arm64), T3 Code Nightly desktop app, Antigravity provider (gemini-4-neon, High)
Additional context (from local conversations/<id>.db, step_type = 17)
Occurred twice in one thread (idx = 18 at 2026-11-20T19:27:40Z and idx = 70 at 2026-11-20T19:30:29Z), both after 10+ successful calls in the same turn:
short_error: NOT_FOUND (code 404): Requested entity was not found.
full_error:
HTTP 404 Not Found
TraceID: 0x5613be76c0af1656 (and 0x7e32de96a1cd3f0c)
Headers: {"Content-Length":["114"],"Content-Type":["text/event-stream"],"Server":["ESF"],...}
{"error":{"code":404,"message":"Requested entity was not found.","status":"NOT_FOUND"}}
Metadata
Metadata
Assignees
Labels
No labels
Activity