What problem does it solve? Multi-agent crews lose track of their own state: which roles are active, whether the codemap and diagrams are stale, which tickets are open, and whether review gates are running. This Skill gives the persistent project-manager agent a single procedure for reading crew state, acting on it within a configured authority level, and keeping the crew's composition and knowledge artifacts current. ## Core Features & Use Cases - Crew state interpretation: Reads the JSON output of crew_state.py (health rate, open tickets, codemap freshness, diagram staleness, incident status, triggers) and reports prioritized findings without re-deriving metrics by hand. - Tiered authority model: Operates under report-only, act, or autonomous modes from .crew/config.json, with hard stops requiring explicit user approval for role offboarding, map deletion, metrics rewrites, and destructive git operations. - Role lifecycle management: Onboards repos and roles with evidence-based justification from .crew/metrics.md, and offboards roles while recording the specific coverage gap left behind. - Use Case: A user asks "how is the crew doing?" — the PM runs crew_state.py, reports that two diagrams are behind HEAD and a handoff note is pending, then under act authority dispatches an explorer and docs-writer to refresh them. ## Quick Start Ask the crew PM to report current crew status and recommend what needs attention.
