You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.
Picks up after /build to publish a GitHub release. Pre-flights the staged artifacts, drafts release_notes.md from commits since the last tag, tags master, then hands the gh release create upload command to the user. Stops short of running the upload — that's user-driven.
Spawn a fresh-context Opus 4.7 agent to give an independent take when Claude and the user are talking past each other on a contested logic interpretation, OR to validate behavioral self-observations from /session-recap Step 0. User-triggered or claude self-proposed at impasse.
End-of-session recap — codify learnings synchronously into their canonical homes; daily logs hold narrative continuity only.
Review dirty code and fix iteratively using Ralph loop pattern. When user say to "loop to fix dirty" or "review+fix"
Review dirty code changes using Claude Code Agent tool. When user say to "review" or "review changes" or "review dirty code"
Sanity-scan the production codebase for cruft that piles up after AI-assisted feature work — duplicated logic, dead helpers, half-finished implementations, speculative architecture, stale comments. Discussion-first, item-by-item; smoke-tests every fix.
Build the JRE-PA-Simulator Windows executable + staged distribution folder (no GitHub release). Embeds version in exe metadata.