Skip to content

Fix stuck Codex working indicator by steering busy messages - #552

Merged
srid merged 4 commits into
masterfrom
better-surge
Sep 6, 2026
Merged

srid merged 4 commits into
masterfrom
better-surge

Conversation

@srid

@srid srid commented Sep 6, 2026 •

Copy link
Copy Markdown
Member

Sending another message while Codex was working opened an overlapping ACP prompt. Older requests could remain unanswered after Codex finished, leaving the panel stuck on “Codex is working…”. Busy sends now use steering when the agent advertises steering without a prompt queue; queue-capable agents retain their existing behavior.

Keep codex-acp 1.10.0 / Codex 0.153.3 and use Nix fetchpatch to backport upstream PR #441, pinned to its commit and content hash. If a turn finishes before steering arrives, the adapter returns the message to Olai to start a tracked prompt. The upstream PR remains open; its branch is based on an older adapter release.

Browser coverage exercises multiple busy sends, completion, cancellation, refused-steering retry, and completion/cancellation races. Chat documentation is updated.

Validation:

  • Full just ci passed on 274a7f483: 49 successful jobs, including all 1,485 browser scenarios.
  • The packaged build includes 10 upstream steering tests.
  • Passed on rerun without code changes; earlier attempts hit intermittent conversation-restoration timeouts.

@srid
srid merged commit f8bfb0d into master Sep 6, 2026
20 checks passed
@srid
srid deleted the better-surge branch September 6, 2026 20:52
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.

1 participant