What problem does it solve? Teams running AI workflow skills accumulate reflection files after each run, but turning that scattered feedback into concrete SKILL.md edits is manual and error-prone. This Skill reads those reflections, finds recurring themes, and produces a structured improvement plan without touching the skills themselves. ## Core Features & Use Cases - Reflection aggregation: Enumerates unarchived reflection files across harness skill roots, parses What worked and Improvements to SKILL.md sections, and gates on a minimum-evidence threshold. - Theme grouping and contradiction surfacing: Groups recurring recommendations by skill, separates actionable changes from speculative notes, and flags contradictory feedback instead of silently resolving it. - Plan generation with handoff: Writes a versioned plan file listing consumed reflections, cross-cutting recommendations, and an archival directive for the downstream skill-editor, plus repo-local handoff state. - Use Case: After several phase-loop runs, invoke the planner to audit accumulated reflections for the roadmap-builder, plan-phase, and execute-phase skills and receive a single plan file the skill-editor can apply. ## Quick Start Ask the assistant to review the accumulated skill reflections and produce an improvement plan for the workflow skills, optionally targeting one skill with --target.
