What problem does it solve? Project harnesses (AGENTS.md, roadmaps, specs, memory, kanban metadata) silently rot over time: dead links, oversized files, stale specs, orphaned worktrees, and stalled epochs accumulate without anyone noticing. This Skill performs a periodic, structured sweep of the current scope, automatically fixing what is unambiguous and surfacing everything else as proposals for human decision. ## Core Features & Use Cases - Script-driven measurement: Runs pop_status.py and pop_validate.py first to gather kanban overview, blocked tasks, pending gates, and limit violations before any manual inspection. - Parallel subagent waves: Delegates collection and correction to waves of 3-5 parallel subagents, one per front (base files, orphaned worktrees, outdated specs, DOX audit, note health, memory, stalled epochs, dated gate debt, swollen modifications, orphaned yolo branches), with disjoint write sets to prevent conflicts. - Strict fix-vs-propose boundary: Directly fixes unambiguous issues in the scope's own harness (dead links, oversized files, completed modification rows) while never touching managed harness files or project content, and proposing anything that changes meaning. - Consolidated report: Writes a structured review report (waiting on you, adjusted, stalled, progress, proposals) into the scope's notes directory, linked from INBOX.md. - Use Case: Ask for a roadmap review at the end of a sprint; the skill sweeps every harness front, reinstalls stale managed harnesses, prunes completed modifications, and hands you a report of decisions only you can make. ## Quick Start Ask the agent to run a weekly review of the current project scope and produce a report of fixes and proposals.
