CI Integration

Add this GitHub Action to your repo. Every pull request that touches Claude Code config files will be checked automatically. Score below threshold = PR blocked.

name: Claude Code Config Quality
on:
  pull_request:
    paths: ['CLAUDE.md', '.claude/**', '.claudeignore']
jobs:
  config-check:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: '22' }
      - run: npx claude-launchpad@latest doctor --min-score 80 --json

The --min-score 80 flag returns exit code 1 if the score is below 80, which blocks the PR. The --json flag outputs structured JSON for programmatic consumption.

The canary pattern (how launchpad guards itself)

Generated configs can rot silently: Claude Code evolves, and a hook that was correct last quarter may no longer fire at all — with no error anywhere. Unit tests can't catch this because they validate what a tool emits, not what Claude Code executes.

Launchpad's own repo runs a weekly canary workflow that scaffolds a fresh project with init -y, installs the latest Claude Code release, and asserts real behavior in live headless sessions:

• the destructive-command hook actually blocks rm -rf /...
• the .env guard actually fires (and the seeded secret stays out of the transcript)
• the PostToolUse auto-format hook actually rewrites a just-written file
• memory install actually registers the MCP server (claude mcp list)
• SessionStart memory context injection actually produces output

Failures open a canary-failure issue automatically. It needs an ANTHROPIC_API_KEY secret and costs a few cents per run (short haiku sessions). If you maintain your own hook library, steal this pattern — it's in scripts/canary.sh.