Skip to content

fix(codex): count reasoning as turn activity - #7

Merged
BenKalsky merged 1 commit into
upgrade/v0190-issue-carriesfrom
fix/codex-reasoning-activity
Jul 30, 2026
Merged

fix(codex): count reasoning as turn activity#7
BenKalsky merged 1 commit into
upgrade/v0190-issue-carriesfrom
fix/codex-reasoning-activity

Conversation

@BenKalsky

Copy link
Copy Markdown
Member

What

Treat completed Codex reasoning items as non-durable turn activity. They remain excluded from final output and message history, but clear the post-tool quiet watchdog.

Why

A live 5b review emitted a completed reasoning item nine seconds after its last tool result, then continued reasoning. Hermes ignored that activity and interrupted the turn at the 90-second post-tool threshold. The hardened oneshot correctly failed closed, but the review was still killed while Codex was active.

Verification

  • 98 projector/session tests passed
  • Ruff passed
  • diff-check passed
  • after merge, rerun the same live review: it must either return completed final evidence or fail for a reason other than post-tool silence after observed reasoning activity

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 70fa0daa45

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".

Comment on lines +101 to +104
# Completed reasoning proves that Codex is still advancing the
# turn. Keep it out of durable messages/final output, but clear
# the post-tool quiet watchdog just like commentary does.
return ProjectionResult(is_activity=True)

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Badge Count streaming reasoning deltas as activity

When Codex spends more than 90 seconds on a single post-tool reasoning item while actively emitting item/reasoning/delta or item/reasoning/summaryDelta, this return is never reached because project() returns an inactive result for every non-item/completed notification at lines 89–90. The post-tool watchdog therefore still interrupts an actively reasoning turn before the completed item arrives; mark reasoning delta notifications as non-durable activity as well so the fix covers long reasoning items, not only reasoning that completes within the timeout.

AGENTS.md reference: AGENTS.md:L54-L57

Useful? React with 👍 / 👎.

@BenKalsky
BenKalsky merged commit 97d6e67 into upgrade/v0190-issue-carries Jul 30, 2026
32 checks passed
@BenKalsky
BenKalsky deleted the fix/codex-reasoning-activity branch July 30, 2026 23:22
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