Native slash commands
/mini:* commands run inside your Claude Code session. mini prints the right prompt; Claude does the work and saves state back — no extra process to babysit.
mini-orchestrator is a CLI that keeps your project moving one verifiable phase at a time — directly inside your Claude Code session.
$ npm install -g mini-orchestrator
Want to evaluate mini without a global install? Run it one-off with npx — it ships with npm, fetches the package into a temporary cache, and runs it once without installing anything globally.
npx mini-orchestrator install-commands
The command asks where to put the /mini:* commands — this project or all projects — and writes only where you choose; it never edits ~/.claude/settings.json on its own. Pick this project to keep everything inside the repo's .claude/commands/mini. When you're done evaluating, remove what it added with mini uninstall.
mini-orchestrator (the mini command) is a small orchestrator that sits on top of Claude Code. It holds your project state — phases and their steps — and drives the work forward in deliberate, reviewable increments instead of one giant leap.
/mini:* commands run inside your Claude Code session. mini prints the right prompt; Claude does the work and saves state back — no extra process to babysit.
Each phase and step lives in .mini/, so progress is explicit, inspectable, and versioned alongside your code.
A single auto-commit per finished phase, an up-to-date project map, and memory notes — generated for you, only when a phase closes.
If you already work in Claude Code, mini-orchestrator adds structure: propose a phase, plan it, implement it, verify it — repeat.
A guided, phase-driven workflow that keeps an AI agent on a short leash and makes every step easy to review before moving on.
Every phase moves through the same four commands. Each one is small enough to finish and verify in a single session.
New to mini-orchestrator? Read the FAQ
Lightweight state in .mini/ tracks where each phase and step stands, with a clear next action at all times.
A machine-generated map across ten languages lets the agent read code targeted by symbol instead of scanning whole files.
Each closed phase leaves a short memory note and a run report — context the next phase can build on.
Optional single auto-commit per phase, version bump and changelog stamp on release, and a soft-reset undo.
/mini:auto chains phases on its own, with a cooperative stop signal whenever you want to step back in.
An optional Claude Code status line shows the working directory, model, and context-window usage at a glance.