Skip to content
Back to skills

Crew Init

ASecurity

Bootstraps the crew scaffold (CLAUDE.md, AGENTS.md, crew/…) on a new or existing project — copies, detects the stack, resolves placeholders. Also triggers on — "/crew-init", "initialise le scaffold crew", "initialize the crew scaffold".

  • 3 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added November 15, 2026
ai-agentspythongoreactgitfrontendsecurity

Works with

  • cli

Security analysis

A100/100

Scanned November 15, 2026

npx -y skills add Nakken13/claude-crew --skill crew-init --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Crew Init?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Crew Init
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/nakken13-crew-init/badge)](https://www.skillsdirectory.com/skills/nakken13-crew-init)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: crew-init
description: Bootstraps the crew scaffold (CLAUDE.md, AGENTS.md, crew/…) on a new or existing project — copies, detects the stack, resolves placeholders. Also triggers on — "/crew-init", "initialise le scaffold crew", "initialize the crew scaffold".
---

This skill runs the bootstrap already described in prose in the scaffold's
`README.md` (`${CLAUDE_PLUGIN_ROOT}/template/`) — it does not replace that
document, it saves running the steps by hand and leaving a placeholder
unresolved by oversight.

## Current-state detection

1. The project already has `CLAUDE.md` + `crew/` filled in (no `<...>` left) →
   do nothing, tell the user the scaffold is already initialized and suggest
   `/crew-status` instead, or `/crew-update` if a newer scaffold version is
   available.
2. The project has `CLAUDE.md` + `crew/` with unresolved `<...>` placeholders →
   resume directly at "Placeholder resolution" below (no need to copy the files
   again).
3. Neither → full bootstrap, steps below from the start.

## Steps

0. **Check the plugin installation**: if `${CLAUDE_PLUGIN_ROOT}` is not defined or
   does not resolve to an existing directory, this skill is running outside a
   plugin context — stop immediately and tell the user to install the plugin
   first (`/plugin marketplace add Nakken13/claude-crew` then
   `/plugin install crew-mates@crew-mates`) before rerunning `/crew-init`. Never
   attempt a partial copy in this case.
1. **Copy** to the project root (without overwriting a file already present and
   already filled in), from `${CLAUDE_PLUGIN_ROOT}/template/`: `CLAUDE.md`,
   `AGENTS.md`, `PRODUCT.md`, `CONTRIBUTING.md`, `SECURITY.md`,
   `check_placeholders.py`, the entire `crew/` folder (empty structure:
   `README.md` + `INDEX.md` per subfolder, no task file).
   **Do not copy** the skills, the agents, or the hook scripts — they run
   directly from the installed plugin (`${CLAUDE_PLUGIN_ROOT}`),
   `/crew-new-task`, `/crew-close-task`, `/crew-status`, `/crew-start` are
   already available with nothing to copy. Touch nothing in
   `.claude/settings.json` for the crew hooks — they are provided by the plugin
   (`hooks/hooks.json`), not by a local entry.
   **Check/install the dependencies of the skills routed in `CLAUDE.md`** —
   `claude plugin list` does not list them all, some are silently missing unless
   checked explicitly:
   - Official plugins (`claude plugin list | grep -E
     "superpowers|security-guidance|playwright|claude-md-management|frontend-design"`):
     for each missing one, `claude plugin marketplace add
     anthropics/claude-plugins-official` (idempotent if already added) then
     `claude plugin install <name>@claude-plugins-official`.
   - `taste-skill` (skill `design-taste-frontend`) — only if the project has a
     front-end stack (detected in the next step): `claude plugin list | grep
     taste-skill`; otherwise `claude plugin marketplace add
     https://github.com/Leonxlnx/taste-skill` then `claude plugin install
     taste-skill@taste-skill`.
   - `graphify` CLI (needed by `graphify init/update/query`, see § graphify of
     `CLAUDE.md`) — check `graphify` is on the PATH; otherwise `pipx install
     graphifyy` (the real pip package name, not `graphify`).
   - `impeccable` and `ui-ux-pro-max` (skills, not plugins — used by the design
     routing) — check `~/.claude/skills/impeccable` and
     `~/.claude/skills/ui-ux-pro-max`; otherwise `git clone
     https://github.com/pbakaus/impeccable.git ~/.claude/skills/impeccable` and
     `git clone https://github.com/nextlevelbuilder/ui-ux-pro-max-skill.git
     ~/.claude/skills/ui-ux-pro-max`.
   - `motion-design-rn`, `accessibility-motion`, `haptics`, `sound-design-ui`
     (only if the project has a React Native/mobile stack) — source not
     identified to date. Check `~/.claude/skills/<name>`; if missing, warn the
     user ("skill not found, source unknown — provide it or remove the RN/mobile
     line from the routing") rather than inventing an installation.
2. **Detect the real stack**: read the manifests present (`package.json`,
   `requirements.txt`/`pyproject.toml`, `go.mod`, `Cargo.toml`, etc.) — never an
   unverified assumption.
3. **Ask via `AskUserQuestion`** what cannot be deduced from the code:
   - Project name, if not already obvious from a manifest.
   - Product vision / target user / scope (`PRODUCT.md`) — option "internal
     tool, not relevant" → reduce the file to one line or delete it.
   - Commit/branch/PR convention (`CONTRIBUTING.md`) — option "no imposed
     convention" → say so explicitly rather than leave a placeholder.
   - Secrets/auth model (`SECURITY.md`) — option "no auth/sensitive data" →
     delete the file.
4. **Placeholder resolution**: replace every `<NOM_PROJET>` and `<...>` in
   `CLAUDE.md`/`AGENTS.md`/`PRODUCT.md`/`CONTRIBUTING.md`/`SECURITY.md` with the
   answers obtained + the detected stack. Repo map and verified commands: only
   commands actually run successfully during this bootstrap, never an untested
   command.
4bis. **Test policy (no extra question)**: the copied `crew/` already contains
   `crew/CLAUDE_CONTEXT/TESTS_POLICY.md`. From the project description (step 3)
   and the code (step 2), fill in the level (`light` by default; `none` only if
   the project is manifestly low-stakes; never `full`) and the risky zones in the
   project's own words. Vague description → `light` with no zones. Do not
   overwrite an already customized file.
5. **Adapt the skills routing**: in the "Skill routing" (skills routing)
   section of `CLAUDE.md`, remove the lines that do not apply to the detected
   stack (e.g. React Native/mobile lines if no RN, `dataviz` if no
   dashboard/chart).
6. If the repo is larger than a few files: run `graphify init .`.
7. Run `python check_placeholders.py` at the root. If placeholders remain,
   resolve them and rerun until exit 0.
8. Ask the user whether to keep `check_placeholders.py` as a pre-commit check or
   delete it (it is a bootstrap tool, not meant to stay by default once the
   scaffold is filled in).
9. **Report**: files created/filled, decisions taken (deleted files such as
   `PRODUCT.md`, lines removed from the skills routing), final status of
   `check_placeholders.py`, and the **deduced test policy** (level + risky zones,
   editable in `TESTS_POLICY.md`).

## What this skill does not do

- Never rewrites an already filled-in `CLAUDE.md`/`AGENTS.md` unless the user
  explicitly asked.
- Does not guess the stack or the commands — reads the manifests / actually runs
  them, never assumes.
- Does not run `graphify init` on a trivial repo (a few files) — not useful at
  that scale.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.