My personal setup for working with AI; what I learned from existing plugins, plus a bunch of my own, built the way I like to work.
Command
Review a GitHub Pull Request — fetch diff via gh, analyze for correctness/bugs, design/patterns, PR hygiene, and security. Present structured findings locally. No GitHub posting.
Command
Triage a PR's unresolved review comments into a checklist, then walk each to one of three outcomes — Act (route the code change to /dispatch or /orchestrate), Skip/defer, or Note (capture a decision for the reply you'll write). Never drafts or posts replies.
Command
Run the pre-push pre-flight checklist before pushing — commits-to-ship summary, branch-name / commit-hygiene / AI-attribution / debug-artifact / hook-bypass checks; stops on any flag.
Command
Prime the session with architectural rules (universal / language / domain / project) relevant to the current task before writing code.
Command
Write an end-of-session recap — request / investigated / learned / completed / next-steps — with propose-confirm-commit flow and optional promotion of Learned items.
Command
Audit code in scope against architectural rules. Report drift findings with propose-confirm-commit per fix and a feedback loop to improve the rule corpus.
Command
Manage the architectural-rule overlay — list / disable / enable / edit / sync / where across shipped / company / user / project tiers.
Command
Interview the user in detail using AskUserQuestion, write a versioned spec under .claude/specs/ /. Specs evolve via versioned files (v1, v2, …); INDEX.md tracks the active version per slug. Use for non-trivial features where requirements need to be surfaced before building.
Command
Alias for /work-state — report a slug's position in the scope chain (envision→spec→draft-plan→blueprint→execute→archive) as deterministic structured state. Read-only.
Command
Append a dated line to the project's canonical CHANGELOG.md (the "what shipped, newest first" record) via propose-confirm — one writer, several callers.
Command
Regenerate .claude/codemap.md for the current project.
Command
Render Mermaid diagrams (L1 module map, L2 file graphs, L3 export index) from .claude/codemap.md to .claude/codemap.diagrams.md and the Obsidian vault.
Command
Report a slug's position in the scope chain (envision→spec→draft-plan→blueprint→execute→archive) as deterministic structured state — per-stage missing/present/stale/done + next-action + the stale-spec-pin check. Read-only.
Command
Session-terminus driver — one /wrap runs the whole closing ceremony in order (coordinate teardown → checkpoint if warranted → recap → close-out → changelog → BACKLOG-shipped-row sweep → build-progress memory), driving the existing organs rather than reimplementing them. Auto-runs the reversible spine; every canonical…
Command
Author a quality test suite for existing code to a standard — detect framework + conventions, propose a confirmable test plan, author via the scope's language-pro agent, characterization-with-flags stance, surface-not-auto-fix on failures.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: