story-kit: done 233_story_auto_assign_backoff_for_failed_merge_items

.story_kit/work/5_done/233_story_auto_assign_backoff_for_failed_merge_items.md

@@ -0,0 +1,22 @@
+---
+name: "Auto-assign backoff for failed merge items"
+---
+
+# Story 233: Auto-assign backoff for failed merge items
+
+## User Story
+
+As a supervisor, I want auto-assign to stop retrying stories in 4_merge/ that have repeatedly failed, so that one broken merge doesn't block the entire queue and burn API credits in an infinite loop.
+
+## Acceptance Criteria
+
+- [ ] When a mergemaster agent completes with gates_passed=false, the story_id is added to a persistent failure set (in-memory is fine, survives across agent completions but not server restarts)
+- [ ] auto_assign_available_work skips items in 4_merge/ that are in the failure set
+- [ ] The failure set is cleared for a story when: a) the story file is manually moved out of 4_merge/, or b) a human explicitly calls start_agent for it
+- [ ] Empty squash merges (merge succeeds but nothing staged) produce a clear, non-retryable error message explaining the feature branch has no unique changes vs master
+- [ ] Regression test: mergemaster fails on story X, auto-assign runs, story X is NOT re-assigned
+- [ ] Regression test: after manual start_agent clears the failure, auto-assign can pick it up again
+
+## Out of Scope
+
+- TBD