Ten commands. One job each.
Each command is a procedure the agent runs literally. Click through for the when and the what it produces.
The ten commands and the CLI fall into four roles around one hub. Knowing which role a piece sits in tells you when it should run.
Each role runs on its own rhythm. Enforcers fire on every commit. Guides fire inside a single change. Authors run when the stack shifts. Orient pieces run on the calendar.
Cadence
The commands are independent. Pick the one you need. Day-to-day, this is when each tends to fire.
| When | What to run |
|---|---|
| Day 0 on a repo | init, then verify |
| Every non-trivial change | plan before code |
| Every real decision | adr |
| Stack or convention shifted | document |
| New area, multi-stack repo | scope |
| Connecting to production | mcp |
| Fresh session or handoff | onboard |
| Quarterly | audit |
All nine
| Command | What it does | Reach for it when |
|---|---|---|
init | Bootstrap a repo with the full context engineering scaffolding. | A new repo, or an existing repo with no CLAUDE.md / AGENTS.md / .cursor/rules yet. |
audit | Score the repo across the five context layers and name the gaps. | "How good is our AI setup?" Or before a migration, a new-hire, or a handoff. |
document | Delta-refresh stale context files without nuking history. | The repo already has CLAUDE.md / AGENTS.md / rules, but they no longer match the code. |
plan | Author a Plan Mode contract before any non-trivial change. | Refactors, migrations, anything touching more than three files or a public API. |
adr | Capture an Architecture Decision Record, including the alternatives you rejected. | A choice was made and the why needs to outlive the conversation. |
scope | Scope rules to file patterns so the right context loads in the right files. | Multiple stacks in one repo, or rules that keep firing where they should not. |
mcp | Design an MCP setup with least-privilege, human-in-the-loop, and real auditing. | Before connecting an agent to anything that can touch production. |
verify | Bake test, lint, and typecheck commands into context so the agent self-corrects. | Every repo. Every time. The single highest-leverage move in this skill. |
onboard | Produce a task-specific orientation brief for a fresh agent or engineer. | New chat session, handoff to a teammate, or compacting a noisy session into a clean restart. |
apply | Apply a remediation plan produced by `audit --html` to the repo. | Right after an interactive audit, to enact the chosen fixes without copy-pasting them by hand. |