Audit
Audit current session's work in parallel (code / spec / test) and apply safe fixes
Checkpoint
Checkpoint - devlog and commit without phase approval
End Session
End session - close devlog, run schema audit + lessons + state updates, optionally suggest slim or playbook extraction, commit.
Start Session
Start session - review last commit and get up to speed
Slim Agent Md
Refactor bloated AGENTS.md or CLAUDE.md files into lean primary instructions plus topic-grouped knowledge files. Agent-neutral alias for slim-claude-md.
Spec Draft
Generate a design specification document from research results and conversation context. Models on the Sentinel-2 spec template (Pick + Why Not blocks, Risk Register, Open Questions, Implementation Kickoff Checklist). Two modes: full (multi-week projects) and lite (2-3 week projects).
Audit Playbooks
Walk the global playbook library, flag stale entries, optionally spawn Explore agents to verify each playbook's claims against current library docs. Run quarterly or before a major project that depends on a playbook.
Slash Goal
Author a SHORT completion-condition for the "slash goal" command that points at the phase doc as the contract and drives it end-to-end — proof-first, bookended by $start-session / $audit / $end-session, stopping only at genuine hard blockers. The phase doc (its pass criteria + Autonomy section) carries the detail; the goal stays terse. Run after /formalize-plan (and optional /plan-audit), before…
Extract Playbook
Distill cross-project knowledge from the current project's agent instructions or knowledge base into a playbook for the global library. Use when you realize a pattern you just solved will recur across projects.
Slim Claude Md
Refactor bloated AGENTS.md or CLAUDE.md files into lean primary instructions plus topic-grouped knowledge files (and where appropriate, new skills). Use when an agent instruction file exceeds 25,000 characters, drifts toward the 40,000 hard ceiling, or has accumulated detailed reference material that belongs in docs/knowledge/.
Find Session
Find and inspect previous Claude Code or Codex session transcripts
Formalize Plan
Formalize a plan into phase docs and implementation plan
Scaffold Project
Create the full development system skeleton for a new project — AGENTS.md, Claude Code compatibility, implementation plan, knowledge base, skills, phases, devlogs, research docs, and git repo. For non-web apps, also scaffolds an MCP co-development surface (Rust MCP server + state tree + ZMQ IPC).
Research Prompt
Create a deep research prompt document for investigating a technical domain. Use when the user wants to explore how to build something, evaluate approaches, or understand a complex technical space before committing to an implementation.
Plan Audit
Pre-implementation four-agent audit on a phase plan (spec-alignment, acceptance-bar/proof-altitude, toolchain-feasibility, sub-phase decomposition). Run after /formalize-plan, before starting work. Catches plan-level issues before code is written — including pass criteria a mock could satisfy — and applies a best-guess fix to the plan doc for every finding, flagging the judgement calls for you to…
Wrap
Lightweight sub-phase close-out — a short devlog plus a commit, then move on. No phase ceremony (no lessons, schema audit, slim/playbook nudges, or Current Status changes). Use when a sub-slice is DONE and you want to roll straight to the next one. For mid-work saves use /checkpoint; for a real phase boundary use /end-session.
Lookup
Search the current project's knowledge base AND the global cross-project playbooks library by keyword. Returns ranked matches with citations. Use when an agent or user wants to recall a project-specific pattern, a cross-project architectural pattern, or a gotcha without reading the whole library.