What problem does it solve? After an epic ships across many coding sessions, no single session saw the whole change, so defects like god-class growth, duplication, and spec drift go unnoticed. This Skill reads the evidence an epic left behind — specs, story files, diffs, commits, and sprint status — and produces a sourced retrospective with an honest acceptance verdict. ## Core Features & Use Cases - Evidence-based analysis: Gathers the epic spec, story files, diff range, and per-story commits via deterministic scripts, then derives aggregate views (architecture delta, duplication map, god-class growth, spec reconciliation) where every finding carries a file, line, or commit reference. - Acceptance verdict with hard rules: Judges the epic against declared or profiled acceptance criteria as accepted, accepted-with-open-items, or rejected; unfinished stories force a machine verdict of rejected unless a human overrides. - Sprint-status integration: Updates sprint-status.yaml atomically through a comment-preserving script that marks the retro done, appends owned action items with stable ids, and validates every write with rollback on failure. - Use Case: After an unattended agent run finishes epic 4, run the retrospective to discover that three stories collectively grew a 3,000-line class, verify the behavior end to end, and record a rejected verdict with routed action items before starting epic 5. ## Quick Start Run a retrospective on the last completed epic and show me the findings with their sources and the acceptance verdict.
