What problem does it solve? Teams running automated PR pipelines often cannot tell how many changes merged in a single pass versus how many needed expensive second passes, or what caused the rework. This Skill reads finished pull requests from the configured tracker, deterministically classifies each run (clean single pass, hard recovery, loop checkpoints, or unexplained), and ranks the recorded causes by the wall-clock hours they cost beyond the median clean run. ## Core Features & Use Cases - Deterministic run classification: A shell classifier (references/classify-runs.sh) parses agent marker comments, reviews, and timestamps to assign each finished PR a class, never relying on model judgement. - Cost-ranked cause report: Causes are ranked by excess hours beyond the clean-run baseline, with coverage disclosures for missing timestamps, sizes, and truncated windows. - Read-only with handoff: The Skill never merges, edits, comments, or labels anything; it offers to pass the top cause to om-prepare-issue for filing. - Use Case: Ask "why is our pipeline slow" and get a report showing that 62% of runs finished in one pass, merge conflicts cost 14 hours beyond baseline across 5 PRs, and 3 second passes have no recorded cause. ## Quick Start Ask the agent to run a pipeline retro over the last 30 days of pull requests and rank what the second passes cost in hours.
