skills

Roadmap triage across the 4N1D framework (NOW/NEXT/NOT-YET/NEVER/DONE in docs/roadmap/). Scans session for completed work, killed ideas, new discoveries, fired triggers, and stale items. Recommends additions, promotions, demotions, and archiving across all five files. Checks anti-patterns (NOW >4, no [learn], stale triggers, zombie NOT-YET). Use at session end, after shipping, or on-demand. Triggers on "/4n1d", "update roadmap", "roadmap triage", "what goes in NOW", "promote to NEXT", "session wrap-up", "what did we ship", or any request to review docs/roadmap/ files.

Audit, improve, and maintain CLAUDE.md (project instructions for AI agents). Detects stale rules, contradictions, missing pitfalls, outdated workspace facts, bloat, and broken file references. Use whenever the user mentions CLAUDE.md, agent instructions, project context, or workspace rules — including "audit my CLAUDE.md", "clean up agent instructions", "add a pitfall", "update project context", "my CLAUDE.md is outdated", "trim CLAUDE.md", or after major codebase changes. Even if the user doesn't name CLAUDE.md explicitly — requests like "why do agents keep getting confused" or "something is wrong with my project setup" may indicate CLAUDE.md needs attention.

Meta-skill that maps cognitive mode-switch words to appropriate thinking behaviors and skills. **AUTO-INVOKE when user uses these trigger patterns:** - Depth: "rigorous", "first-principles", "mechanistic", "axiomatic", "falsifiable", "steelman", "non-trivial", "exhaustive", "load-bearing", "dispositive", "ultra think hard" - Reframe: "reimagine", "orthogonal", "contrarian", "subversive", "invert", "counterfactual", "zero-based", "provocative", "transmute", "reframe" - Precision: "probabilities", "bounded", "monotonic", "asymptotic", "invariant", "order of magnitude", "sensitivity analysis", "convex", "distribution", "marginal", "dimensionality" - Tradeoff: "sacrifice", "irreversible", "binding constraint", "pareto", "second-order", "opportunity cost", "regret minimization", "non-negotiable", "asymmetric" - Quality: "elegant", "idiomatic", "canonical", "production-grade", "battle-tested", "minimalist", "composable", "tasteful", "crisp" - Systems: "emergent", "cascading", "equilibrium", "phase transition", "fee

Retroactively analyze exported sessions to reveal orchestration that wasn't captured. Use --reconstruct for old sessions where you forgot --showcase. Infers skill logic, agent internals, and decision rationale from transcript patterns with 60-80% accuracy.

Capture full orchestration details (skills, agents, decisions, compound learning) in your session transcript for YC or investor demos. Use --showcase flag at session start. Works with Claude's built-in /export command.