Repository navigation
dev-lead retry sweep posts blank-looking PR comments: dispatch guards and retry markers are marker-only bodies that stay on the PR #2200
Description
Problem
scripts/dev-lead-retry.sh records its state as PR comments whose whole body is an HTML comment. GitHub renders those as empty comment boxes. A stalled PR collects one per retry attempt and per dispatch, so maintainers see a stream of "empty comments" from dev-lead.
Evidence
| Comment | Body built at | Posted at | Lifecycle |
|---|---|---|---|
Dispatch guard: <!-- dev-lead-dispatch-guard sha=… at=… --> |
post_dispatch_guard, dev-lead-retry.sh:312-319 |
:595, :706, :727 |
Never deleted |
Bot-comment retry marker: <!-- dev-lead-bot-comment-retry id=… version=… attempt=… at=… --> |
bcr_retry_marker, scripts/lib/bot-comment-retry.sh:292-295 |
:943 |
Withdrawn only when the dispatch is skipped or fails; stays forever after a successful dispatch |
Bot-thread retry marker: <!-- dev-lead-bot-thread-retry threads=… attempt=… at=… --> |
btr_retry_marker, scripts/lib/bot-thread-retry.sh:217-219 |
:1174 |
Same as the bot-comment marker |
- The two retry markers are recent: A lost fix-bot-comment run is never retried — a bot comment stays undispositioned and the PR stalls at the maintainer-comment gate forever (#2009) #2017 (2026-10-03) and dev-lead: unprocessed bot review threads are never retried — review-thread counterpart of #2017 #2046 (2026-10-08). That matches when the blank comments started to be noticed.
- Every other script-posted dev-lead comment has visible text after its marker. These three are the only marker-only bodies.
Acceptance criteria
- No dev-lead comment renders blank. Each of the three bodies gets one short visible line after the marker, for example
_dev-lead: retry dispatched for review thread(s) (attempt 2/3)._. Readers that match these markers keep working, since they match the marker, not the whole body; verify each reader, including the maintainer-comment gate, which must still treat them as automation. - Spent markers stop piling up. Either minimize them (
OUTDATED) once the attempt they guard has finished or expired, or move this dedup state off the PR timeline (for example to a single edited status comment per PR). - Tests. A bats case asserts every body posted by
dev-lead-retry.shhas non-marker visible text.
Related
- dev-lead retry sweep ignores dev-lead:hands-off: it posts retry markers and dispatch guards and dispatches passes on held PRs #2089 (stopped these markers on held PRs).
- dev-lead: suppress visible body on no-changes terminal comments #362 (earlier "suppress visible body" decision for no-changes comments, which is the opposite trade-off).
- dev-lead recovery can queue a duplicate fix-reviews run after the 600s dispatch guard expires — use a durable in-flight claim #1797 (the dispatch guard's design).
Filed from a Claude Code session auditing dev-lead gaps, 2026-10-10.
Activity
github-actions commentedon Oct 10, 2026
Thanks for the detailed report and the line-level evidence, which makes this easy to follow. The acceptance criteria are clear, so I have one question about the second one. Should spent markers be minimized (
OUTDATED) once their attempt finishes or expires, or should the dedup state move into a single edited status comment per PR? The first option is a smaller change. The second removes the timeline noise entirely but means reworking how the readers find that state.don-petry commentedon Oct 10, 2026
Two more examples, both from the bot-comment retry marker, on PR #2205 within 10 minutes of each other:
<!-- dev-lead-bot-comment-retry id=IC_kwDOR9SdIs8AAAABa5W5ig version=2026-10-10T17:15:41Z attempt=1 at=2026-10-10T17:40:11Z --><!-- dev-lead-bot-comment-retry id=IC_kwDOR9SdIs8AAAABa5YW1Q version=2026-10-10T16:58:23Z attempt=1 at=2026-10-10T17:50:37Z -->Both were posted by
donpetry-botand render as empty comment boxes on the PR.The first is retrying CodeRabbit's "Review limit reached" comment. That notice is edited in place every few minutes while the rate limit counts down, and each edit gets a new
version. If nothing filters these out, a rate-limited PR can collect a fresh blank retry marker every time CodeRabbit updates its countdown.Generated by Claude Code