fix: rework pickup ignores feedback when the verdict was COMMENT (reassignment starves the agent of context) #476

New issue

Closed

opened 2026-08-09 02:05:06 +00:00 by igor · 0 comments

igor commented

2026-08-09 02:05:06 +00:00

Collaborator

Known gap, now with two field failures (igor#468, 2026-08-09): when a PR is REASSIGNED to the bot while its latest shadow verdict is COMMENT (not REQUEST_CHANGES), the rework agent receives no feedback payload -- .review.pending_rc_body is only populated on an RC verdict -- so it sees nothing actionable and exits without commits. Both rework rounds on #468 stalled exactly this way while an explicit, actionable reviewer comment sat on the thread unread.

The change: in the PR-reassignment pickup path (bin/tick.sh / wherever the rework user message is assembled), when there is no pending RC body, include the recent PR discussion instead: the last shadow-review comment AND any human/operator comments posted after the latest review marker (). Fence them as feedback to address, same framing as the RC body gets.

Guard: keep the existing no-op path -- if there are genuinely no comments beyond the reviewed state, the current "no changes made" exit is correct.

Tests: extend the relevant bin/test-*.sh (review/rework surface): (1) reassignment with COMMENT verdict + a post-review operator comment -> the assembled prompt contains that comment; (2) reassignment with no post-review comments -> prompt assembly unchanged from today.

Known gap, now with two field failures (igor#468, 2026-08-09): when a PR is REASSIGNED to the bot while its latest shadow verdict is `COMMENT` (not `REQUEST_CHANGES`), the rework agent receives no feedback payload -- `.review.pending_rc_body` is only populated on an RC verdict -- so it sees nothing actionable and exits without commits. Both rework rounds on #468 stalled exactly this way while an explicit, actionable reviewer comment sat on the thread unread. **The change:** in the PR-reassignment pickup path (`bin/tick.sh` / wherever the rework user message is assembled), when there is no pending RC body, include the recent PR discussion instead: the last shadow-review comment AND any human/operator comments posted after the latest review marker (``). Fence them as feedback to address, same framing as the RC body gets. **Guard:** keep the existing no-op path -- if there are genuinely no comments beyond the reviewed state, the current "no changes made" exit is correct. **Tests:** extend the relevant `bin/test-*.sh` (review/rework surface): (1) reassignment with COMMENT verdict + a post-review operator comment -> the assembled prompt contains that comment; (2) reassignment with no post-review comments -> prompt assembly unchanged from today.

igor added the

Agent

label

2026-08-09 02:05:06 +00:00

igor self-assigned this

2026-08-09 02:59:28 +00:00

igor referenced this issue from a pull request that will close it,

2026-08-09 03:10:28 +00:00

fix: surface the shadow review's own comment on a plain PR reassignment #479

igor added spent time

2026-08-09 03:10:28 +00:00

10 minutes 32 seconds