fix: TaskOutput-timeout guard, actor-permission preflight, report deps - #7
Merged
Conversation
Follow-ups from auditing dashpay/dash-evo-tool#908's successful run (29744073162) after the run_in_background fix landed: - TaskOutput can time out mid-block and return status:running instead of a final result; the prompt now requires re-blocking on timeout instead of treating that as "collected". Made explicit that this rule is transitive — the orchestrator must pass it down to any specialist it spawns (e.g. via grumpy-review), since a nested reviewer that itself fans out sub-agents has no way to know it's running headless otherwise. Observed live: a QA sub-agent hit exactly this and only recovered by model discretion, not enforcement. - Add a fast write-access preflight: the underlying claude-code-action rejects non-write-access triggering actors from deep inside its own execution (~40s in, opaque "Action failed with error" message). Check via `gh api .../collaborators/{actor}/permission` before paying that cost. - Install claudius's documented report-rendering Python deps (markdown, jsonschema, jinja2, etc.) via apt — the runner image doesn't have them, which was crashing generate_review_report.py mid-run (`import markdown`) and getting silently retried around. - Surface (rather than swallow) trigger-label removal failures — a silent failure there leaves the label on the PR and re-triggers a full review on the next push.
1 task
3 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Follow-ups from auditing dashpay/dash-evo-tool#908's first successful run (29744073162) after the run_in_background fix (#6) landed. The fix worked mechanically, but the audit surfaced a few more things worth hardening:
TaskOutputtimeout re-block guard.TaskOutput(block=true)can time out and returnstatus: runninginstead of a final result — that's not completion. The prompt now requires re-blocking on timeout instead of moving on. Made this rule explicitly transitive: the orchestrator must pass it down to any specialist it spawns (e.g. viagrumpy-review), since a nested reviewer that fans out its own sub-agents has no way to know it's running headless otherwise. This wasn't theoretical — the QA specialist in that same run hit exactly this at the 10-minute mark and only recovered by model discretion, not enforcement.claude-code-actionrejects triggering actors without write access from deep inside its own execution (~40s in, an opaque "Action failed with error: User does not have write access on this repository"). We hit this for real on the first re-trigger attempt. Added a fastgh api .../collaborators/{actor}/permissioncheck up front so it fails in a few seconds with a clear message instead.generate_review_report.pycrashed onimport markdownmid-run (the CI runner image doesn't have claudius's documented Python deps) and got silently retried around. Installs them via apt before the review runs.2>/dev/null || true— a silent failure there leaves the trigger label on the PR, which re-triggers a full review on the next push. Now surfaces a::warning::.Test plan
python3 -c "import yaml; yaml.safe_load(open('action.yml'))"— YAML parses cleanly (yamllintunavailable locally)uses: lklimek/claudius-review-action@fix/ci-run-followups— in particular confirm the write-access preflight fails fast and cleanly for a non-write actor, and that report generation no longer hits the markdown import error🤖 Generated with Claude Code