fix(delegation): require progress evidence before retiring subagents - #179
thisisjun786 wants to merge 2 commits into
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![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.
Reviewed commit: 3767ae8826
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".
| copy (packet-failure reclaim, 260711 fork-debate verdict #4); under configured | ||
| first fallback the protocol result owns that decision — `main-direct` permits | ||
| reclaim, `reconcile`/`stop` permit neither reclaim nor a replacement spawn. Lineage: |
There was a problem hiding this comment.
When configured first fallback returns reconcile or stop after two agent failures, this new rule forbids main from reclaiming the work, but plugins/codexclaw/skills/pabcd/references/plan-output.md:18 still requires every delegation plan to state unconditionally that main reclaims after two failures. A coordinator following that template can therefore start direct work while execution remains unreconciled, potentially overlapping an existing writer; update the template to defer to the managed protocol result.
Useful? React with 👍 / 👎.
1 participant
Closes #178.
An executor can keep investigating while its working tree stays clean. The old
waiting guidance treated roughly three empty waits as dispatch failure, while
the delegation reference described wait timeout as a normal outcome.
This change makes the waiting reference the decision owner: compare meaningful
progress, request a non-interrupting checkpoint when needed, and confirm
stagnation before retirement. It preserves explicit cancellation and resource
limits, checks actual shutdown before handing off work, and keeps replacement
and main reclaim behind the existing managed dispatch results.
This is agent-followed guidance, not a runtime watchdog. Automated checks cover
document hygiene and owner routes; a separate reviewer evaluates the decision
procedure against the recorded scenarios. No model-compliance guarantee is made.
Verification at
3767ae88:npm run gatepassed.git diff --checkpassed.The scenario review covers active analysis without edits, long commands, repeated busywork, missing visibility, queued checkpoints, cancellation/limits, and shutdown/recovery authority. Local checks did not rerun unrelated runtime suites; hosted CI reports those separately.
Verification record.