Skip to content

fix(ci): drop --full-auto arg the codex CLI no longer accepts - #83

Merged
PetrefiedThunder merged 1 commit into
mainfrom
claude/production-sweep-cors-fix-7qw61i
Aug 11, 2026
Merged

fix(ci): drop --full-auto arg the codex CLI no longer accepts#83
PetrefiedThunder merged 1 commit into
mainfrom
claude/production-sweep-cors-fix-7qw61i

Conversation

@PetrefiedThunder

Copy link
Copy Markdown
Collaborator

Summary

  • Remove codex-args: '["--full-auto"]' from .github/workflows/codex-autopilot.yml — one deleted line.

Why

Codex Autopilot has failed on every scheduled run that reached the Codex step since 2026-08-07 (run 31191150738, run 31402707055). The failing step's logs show:

error: unexpected argument '--full-auto' found
Error: codex exited with code 2

The workflow uses the unpinned openai/codex-action@v1 tag, which installs the latest codex CLI. A CLI release between Aug 5 (last green run, same repo commit e7a4855) and Aug 7 (first arg-rejection failure, identical commit and workflow) removed --full-auto from codex exec. Since nothing in the repo changed between the green and red runs, the regression is in the CLI, not this repo.

The flag is redundant here anyway: the action already passes sandbox: workspace-write, which is what --full-auto implied.

Note: the workflow's "Explain Codex Action failure" step blames OPENAI_API_KEY for this class of failure; the key was present and never exercised — don't be misled by that line in past logs.

Verification

  • Failing step env in both red runs shows CODEX_ARGS: ["--full-auto"] as the source of the rejected argument; baseline pytest and the secret-validation step passed before the crash in both.
  • The Aug 5 success / Aug 7 failure ran the identical commit and workflow, isolating the change to the unpinned CLI.

Alternative considered

Pinning the codex CLI/action version would also fix this and would prevent future drift, but it's a policy choice about how this workflow tracks upstream — left to the maintainers. This PR is the minimal change that restores the scheduled runs.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Vi3SmxxDFvyF9t5XE8iMm9


Generated by Claude Code

Codex Autopilot has failed on every scheduled run since 2026-08-07
(runs 31191150738, 31402707055): the unpinned openai/codex-action@v1
installs the latest codex CLI, and a release between Aug 5 and Aug 7
removed the --full-auto flag from `codex exec`, so the step exits
immediately with "error: unexpected argument '--full-auto' found"
(exit code 2). The same commit and workflow succeeded on Aug 5, so the
regression is in the CLI, not the repo.

The flag is redundant here: the action already runs with
sandbox: workspace-write, which is what --full-auto implied. Drop the
codex-args input entirely.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vi3SmxxDFvyF9t5XE8iMm9
@PetrefiedThunder
PetrefiedThunder marked this pull request as ready for review August 11, 2026 09:33
@PetrefiedThunder
PetrefiedThunder merged commit c5a7cab into main Aug 11, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants