Expose auto-fix via MCP and improve settings#10
Conversation
aandersen2323
deleted the
codex/expose-auto-fix-endpoint-via-mcp-bridge
branch
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if final_status == "apply_fix": | ||
| fix_req = _build_auto_fix_fix_request(base_req, auto_req, parsed_analysis) | ||
| print(f"Running fix infer (task: {fix_req.task})...") | ||
| fix_result = run_inference(fix_req) |
There was a problem hiding this comment.
The new auto‑fix loop only executes the fix step when the analysis result explicitly sets status to "apply_fix". Previously, a fix run was triggered whenever the analysis supplied fix_instructions, even if the status string could not be parsed. LLM outputs often include valid instructions while returning an unexpected status, so this regression will cause the workflow to stop after the analysis phase and never apply the suggested repair. Consider running the fix whenever actionable instructions are present, regardless of whether the status is recognized.
Useful? React with 👍 / 👎.
1 participant
Summary
/infer/auto-fixAPI through the MCP bridge so remote clients can trigger repairspydantic_settingsfallback and refresh tests to cover the new behaviorTesting
https://chatgpt.com/codex/tasks/task_e_68faf8131e38833198346d5de7b03785