mise-en-claude installs Claude Code prompts to set up or rescue a project with git, CLAUDE.md, structure, connectors, CLI tools, and workflow. It supports Mode A and Mode B for setup and rescue workflows.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
mise is a paste-in prompt for Claude Code that interviews you about your project and then stands up (or upgrades) a real engineering foundation for it: git, a CLAUDE.md, a project structure, connectors and CLI tools, workflow skills like /mise-cook, and a task tracker. It operates in consent-first, snapshot-safe modes and includes safeguards for secrets by reference.
How it works
The repository provides a CLI/install flow to place Claude Code workflow commands into a project, either via Mode A (zero install) or Mode B (clone + install). It sets up phases from Vision to Verify + hand off, configuring git, context, connections, CLI tools, skills, persistence, and rescue/update behavior. It includes Rescue mode for existing projects and Update mode for reconciliation.
Getting started
Mode A: zero install (fastest)
- Start a Claude Code session for your project (new or existing):
- CLI: open a terminal in the project folder and run
claude. - Desktop app: open Claude Code and open/select your project folder.
- Web / IDE: open your project in claude.ai/code or your IDE extension.
- CLI: open a terminal in the project folder and run
- Copy the contents of
PROMPT.mdand paste it as your first message.
Read https://raw.githubusercontent.com/emadd/mise-en-claude/main/PROMPT.md and follow it to set up my project.
Mode B: clone + install (the workflow commands)
Clone the repo and run the installer to put mise's workflow commands into your Claude Code config:
git clone https://github.com/emadd/mise-en-claude
cd mise-en-claude && ./install.sh # installs /mise-cook, /mise-handoff + /mise-clean for all your projects
# or: ./install.sh --project . # install into just this project's ./.claude
That gives you /mise-cook, /mise-handoff, and /mise-clean, plus the orchestration playbook, globally or per-project. The install is non-destructive and idempotent.
What it sets up
mise works in consent-gated phases from Vision to Verify + hand off, creating a starter project structure, git settings, a CLAUDE.md, connections, CLI tools, and a task tracker integration. It supports Rescue mode for existing projects and Update mode for keeping a project in sync with guidance.
Started on the wrong foot? Rescue mode
If an existing project is detected, it switches to rescue: snapshot first, health report, triage with safety-first fixes, and preserve work. It can write a CLAUDE.md read from existing code and handles committed secrets with guidance to rotate keys without rewriting history behind the user’s back.
Keeping it current: Update mode
Update mode fetches the latest canonical mise, snapshots first, then applies non-destructive updates after user approval. It records a small .mise/ stamp to track what was configured.
Safety: consent-first by design
The tool previews the full plan, asks before each phase, merges or appends files rather than overwriting silently, supports dry runs, and explains each step. It uses secrets-by-reference and ensures plaintext does not enter the model context.
Proven on itself
The repository contains a QA harness under tests/ with a safety rubric, a playbook, and synthetic fixtures to validate behavior across models. It emphasizes snapshot, verify, and review-before-ship discipline for the tool itself.
Supported stacks
Stacks are defined in stacks/ with frontend/mobile options (ios-swiftui, react, nextjs, angular, flutter, react-native) and backend options (node-express, python-fastapi, python-django, generic).