What problem does it solve? AI coding agents forget everything between sessions, so the same mistakes repeat and hard-won patterns are lost. This Skill turns scattered agent session logs into a persistent, cross-agent procedural memory system with distilled rules, confidence tracking, and safety guards. ## Core Features & Use Cases - Context Hydration: Run cm context "<task>" --json before any non-trivial task to retrieve relevant playbook rules, anti-patterns, and history snippets scored by relevance. - Playbook Learning: Rules are distilled from sessions via an automated reflect-and-curate pipeline, with confidence decay, maturity progression (candidate → established → proven), and anti-pattern inversion. - Trauma Guard: Blocks dangerous commands (e.g., rm -rf /, DROP TABLE, force pushes) via Claude Code and Git pre-commit hooks based on learned incident patterns. - Use Case: A pattern discovered during a Cursor session automatically surfaces as a rule when Claude Code starts a similar task the next day, with feedback marks (cm mark <id> --helpful/--harmful) continuously refining rule quality. ## Quick Start Ask the agent to run cm context with a description of your current task in JSON mode, then follow the returned rules and leave inline feedback comments as you work.
