What problem does it solve? Autonomous agent workflows accumulate failures, rework, and misleading measurements that no one systematically reviews. This Skill closes that loop by auditing normalized session history and durable outcomes, then filing evidence-backed improvement tasks so process problems get fixed instead of repeating. ## Core Features & Use Cases - Run History Review: Discovers and loads agent sessions through a normalized history interface, reconciling transcript claims against task state, git history, PR checks, and what actually landed. - Evidence-Based Findings: Publishes findings only when supported by multiple occurrences or one severe case, with causal mechanisms, counterevidence, and testable acceptance criteria. - Governed Publishing: Writes dated audit reports under docs/audits and files inputRequired tasks under docs/plans on an isolated branch, without ever modifying governance, roadmap, or merging work. - Use Case: After a month of autonomous runs, invoke this Skill to detect recurring verification failures, produce a Kaizen report, and open a PR proposing maintainer decisions with measurable acceptance criteria. ## Quick Start Ask the agent to run a Kaizen review of the last 30 days of autonomy sessions and publish a report with any findings as inputRequired tasks.
