Bootstrap Agentic Repo

Bootstrap Agentic Repo

One-time setup that turns any repo into a fully-equipped agentic workspace. Installs skills, runs setup wizards, and verifies the result.

Phase 1 -- Assess

• Detect primary language(s) from file extensions and config files
• Detect frameworks (React, Next.js, Django, Rails, etc.)
• Check for existing CI setup (.github/workflows/, .gitlab-ci.yml, etc.)
• Check for existing skills (ls ~/.claude/skills/ or .skills/)
• Check for existing CLAUDE.md and note what's already configured
• Present findings to user and ask them to confirm or correct

If detection is ambiguous or the repo is barebones (no config files, no clear framework), use AskUserQuestion to ask:

• What language(s) does this project use?
• What framework(s) if any?
• What test framework and runner?
• What CI platform (GitHub Actions, GitLab CI, none)?

Never guess -- always confirm with the user when uncertain.

Phase 2 -- Install skills

Install the internal skills repo and each external skill source:

# Internal skills (this repo)
npx skills@latest add mohganji/skills

# External skill sources
npx skills@latest add mattpocock/skills
npx skills@latest add browser-use/browser-harness
npx skills@latest add remotion-dev/remotion

Only install stack-specific external skills if the stack was detected or confirmed in Phase 1:

# Only if project uses React
npx skills@latest add millionco/react-doctor

• Run each install command
• Report any failures and continue with remaining installs
• List all successfully installed skills

Phase 3 -- Run setup skills

Based on the stack detected in Phase 1, interactively offer one-time setup skills. Each setup skill owns its own CLAUDE.md section -- bootstrap does not write CLAUDE.md directly.

Use AskUserQuestion to let the user pick which setup skills to run:

1. setup-caveman -- Set caveman as default communication style
2. setup-crap-check -- Add CRAP score enforcement via CI and pre-commit hook
3. setup-dry -- Add DRY violation detection via CI and pre-commit hook
4. setup-react-doctor -- Add React performance analysis via CI and pre-commit hook (if project uses React)
5. setup-remotion -- Scaffold remotion-videos/ directory and add Remotion conventions to CLAUDE.md
6. setup-matt-pocock-skills -- Configure issue tracker, triage labels, domain docs (from mattpocock/skills)

Run each selected setup skill in order. Wait for each to complete before offering the next.

Phase 4 -- Verify and summarize

• List all installed skills with ls ~/.claude/skills/*/SKILL.md
• Show what was configured by setup skills (CLAUDE.md sections, CI workflows, etc.)
• Suggest next steps:
  • Try a skill: /cut-the-crap, /tdd, /caveman
  • Read the skills README for the full catalog
  • Run /write-a-skill to create custom skills for this project

Key rules

• Never modify CLAUDE.md directly -- defer to setup skills
• Always show the user what will be installed before running installs
• If a skill is already installed, skip it and note that it's current
• If an install fails, continue with the rest and report failures at the end
• This skill is idempotent -- running it again skips already-installed skills