Review changes on current branch and output a structured review. Optionally post to PR with --pr flag, or apply non-blocking follow-ups inline with --apply-followups.
Run an interactive code-only review for a gza task's implementation branch and produce structured review output compatible with gza-task-improve
Check the implementation against the behavior specs in specs/behavior/. Reports where the code diverges from intended behavior — each divergence is either a code bug or a spec gap. The behavior spec is the source of truth; this skill never edits code or the spec.
Check the behavior spec set for coherence, ownership boundaries, and plain-language discipline without editing the spec or the code
Turn the recurring `watch` stuck-task pile into (1) a diagnosis of why each class is stuck, (2) the existing stuck rows actually cleared now, and (3) systemic prevention so it does not recur. Snapshots watch/incomplete/queue, buckets stuck tasks by failure class, dedups against already-tracked `system` work, unsticks each row by its clearing action (drop moot/dead/stale, spawn follow-up, hand review-loop rows to /gza-task-fix), then ranks and files `system`-tagged prevention fixes by blast radius (cascade-preventer first). Never merges, retries, resumes, deletes branches, or edits code.
Triage `gza incomplete` rows — classify each unresolved merge-unit lineage and recommend the right corrective action (drop moot leaves, escalate to fix, surface manual-resolve rebases, etc.). Never merges, retries, resumes, or deletes branches; never edits code.
Diagnose why a gza task failed — analyzes logs, detects loops, checks diffs, compares baselines, and suggests fixes
Synthesize operator triage guidance from failed-task history, unimplemented plan and explore work, unmerged branches, and queue state