Repository navigation
fix: preserve primary inference failures and rejection causes - #42
Merged
Merged
Conversation
AppSolves
enabled auto-merge (squash)
November 15, 2026 17:44
AppSolves
disabled auto-merge
November 15, 2026 17:44
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The proxy could overwrite a precise upstream error with a cleanup/pipeline error and falsely report provider_stream_failed when Codex closed HTTP after a valid response.completed. Engine output normalization also hid why an otherwise completed turn could not become a valid review.
Preserve a bounded per-request start/finish record with independent primary/cleanup failures, HTTP/SSE metadata and allowlisted transport codes. Success clears recovered errors; older requests cannot replace a newer one. The Docker host forwards only sanitized recent failures and final request state. Add fixed engine rejection reasons without exposing model text. All failed/incomplete analysis still fails CI and remains non-approving.
Validation: npm run check (216 tests), clean checkout, workerd/D1, browser suite, model catalog, npm audit (zero vulnerabilities), Actionlint, Wrangler dry-run, Docker build and actual pinned Codex sandbox. Native regressions cover clean/incomplete responses whose HTTP body stays open and schema/JSON-invalid output. Provider responses are synthetic; no live inference.
No prompt, model, inference timeout, context budget, authorization, credential, quota or merge-policy changes. The source config temporarily closes the merge gate for the established immutable pin rollout; restore it only after caller/Worker coordination. No database migration or new production review is required.