What happened
PR #1348 added diff-hunk validation to findingsToReviewComments in internal/cli/postreview.go, which prevents the 422 errors described in #1067. The fix validates inline comment line numbers against parsed @@ ... @@ hunk headers before submission, and falls back gracefully (skips line filtering) for binary/truncated patches with empty hunks. However, #1067 remains open with the ready-to-code label, meaning a code agent could be dispatched to implement a fix that already exists.
What could go better
When a human fixes an issue outside the normal agent workflow (i.e., the PR branch is not agent/{issue}-{slug}), there is no automation linking the PR back to the issue for closure. The human author didn't reference #1067 in the PR (only the PR description mentions the symptom, not the issue number). This creates a risk of redundant work: a code agent could be dispatched on #1067 and spend tokens reimplementing something that already shipped. Confidence: high — the PR directly addresses the problem described in #1067.
Proposed change
Close issue #1067 as fixed by PR #1348. Additionally, consider adding guidance to CONTRIBUTING.md or the PR template reminding authors to reference related issues with Fixes #NNN or Closes #NNN so GitHub auto-closes them on merge. This is a lightweight process improvement that would prevent orphaned issues after human-authored fixes.
Validation criteria
Issue #1067 is closed. No code agent run is dispatched against #1067 after this retro.
Generated by retro agent from #1348
What happened
PR #1348 added diff-hunk validation to
findingsToReviewCommentsininternal/cli/postreview.go, which prevents the 422 errors described in #1067. The fix validates inline comment line numbers against parsed@@ ... @@hunk headers before submission, and falls back gracefully (skips line filtering) for binary/truncated patches with empty hunks. However, #1067 remains open with theready-to-codelabel, meaning a code agent could be dispatched to implement a fix that already exists.What could go better
When a human fixes an issue outside the normal agent workflow (i.e., the PR branch is not
agent/{issue}-{slug}), there is no automation linking the PR back to the issue for closure. The human author didn't reference #1067 in the PR (only the PR description mentions the symptom, not the issue number). This creates a risk of redundant work: a code agent could be dispatched on #1067 and spend tokens reimplementing something that already shipped. Confidence: high — the PR directly addresses the problem described in #1067.Proposed change
Close issue #1067 as fixed by PR #1348. Additionally, consider adding guidance to CONTRIBUTING.md or the PR template reminding authors to reference related issues with
Fixes #NNNorCloses #NNNso GitHub auto-closes them on merge. This is a lightweight process improvement that would prevent orphaned issues after human-authored fixes.Validation criteria
Issue #1067 is closed. No code agent run is dispatched against #1067 after this retro.
Generated by retro agent from #1348