What problem does it solve? Insights about how AI skills could be improved surface during real work but get lost between sessions. This Skill runs a retrospective pass over one finished project version, mining the whole version's session history for repeated corrections, workflow patterns, and methodology gaps worth turning into new skills or fixes to existing ones. ## Core Features & Use Cases - Version-level retrospective: Run once per project version, right after project-retro, to review everything that happened rather than monitoring per-session. - Structured observation log: Captures findings as numbered observations with Status, Issue, Improvement, and Principle fields in a persistent workspace log, with concurrency-safe numbering and archival rules. - Weekly review workflow: A scheduled or 7-day fallback review cross-checks OPEN observations against all skills, applies non-escalated improvements, and stages updated skills for user approval. - Use Case: After finishing v1 of a project, invoke it explicitly to surface that users repeatedly corrected the same report-formatting rule, then turn that pattern into a staged update to the reporting skill before starting v2. ## Quick Start Run task-observer to review this finished project version and tell me what skill improvements came out of it.
