What problem does it solve? AI agent skills stay frozen after creation: corrections, preferences, and workflow insights discovered during real work are lost between sessions, and nobody systematically identifies which new skills should be built. This Skill runs alongside every work session, captures those signals as structured observation files, and drives a review cycle that turns them into skill improvements. ## Core Features & Use Cases - Continuous observation logging: Watches tool-using sessions for user corrections, recurring patterns, and methodology gaps, writing one Markdown file with YAML frontmatter per observation into a persistent observation-log directory. - Skill candidate discovery: Flags repeating workflows that no existing skill covers and proposes new skills, validating targets and checking sibling skill families before writing. - Scheduled review cycle: Tracks a last-review-date file, offers or runs a comprehensive review of open observations, and stages approved skill updates in a skill-updates folder without editing live skills. - Use Case: A consultant with 30 installed skills notices the agent keeps making the same report-formatting mistake. The observer logs the correction, a later review confirms the pattern, and a staged update to the reporting skill is presented for approval. ## Quick Start Install the skill folder with its references and scripts, add the activation block to your CLAUDE.md, then ask the agent at the end of a work session: any observations logged?
