CI/CD & Automation

Claude Code can run non-interactively in CI/CD pipelines. This enables automated code review, PR descriptions, migrations, and more.

Headless Mode Basics

Two flags make Claude non-interactive:

# -p: Single prompt, exits after response
claude -p "explain this codebase"

# --print: Output only (no interactive UI)
claude --print -p "what does this function do"

Combine them for CI:

claude --print -p "review the changes in this PR for security issues"

Environment Setup

Claude needs an API key in CI:

# GitHub Actions example
env:
  ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}

For headless operation, also set:

# Accept all permissions (use carefully)
claude --dangerously-skip-permissions -p "..."

# Or pre-approve specific tools in settings

GitHub Actions Examples

Automated PR Review

name: Claude PR Review

on:
  pull_request:
    types: [opened, synchronize]

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - name: Install Claude Code
        run: npm install -g @anthropic-ai/claude-code

      - name: Review PR
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
        run: |
          claude --print -p "
            Review the changes in this PR.
            Focus on: bugs, security issues, performance.
            Be concise. Output as markdown.

            $(git diff origin/main...HEAD)
          " > review.md

      - name: Post Review Comment
        uses: actions/github-script@v7
        with:
          script: |
            const fs = require('fs');
            const review = fs.readFileSync('review.md', 'utf8');
            github.rest.issues.createComment({
              owner: context.repo.owner,
              repo: context.repo.repo,
              issue_number: context.issue.number,
              body: review
            });

Generate PR Description

name: Generate PR Description

on:
  pull_request:
    types: [opened]

jobs:
  describe:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - name: Install Claude Code
        run: npm install -g @anthropic-ai/claude-code

      - name: Generate Description
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
        run: |
          claude --print -p "
            Generate a PR description for these changes.
            Include: summary, key changes, testing notes.

            $(git log origin/main...HEAD --oneline)
            $(git diff origin/main...HEAD --stat)
          " > description.md

      - name: Update PR
        uses: actions/github-script@v7
        with:
          script: |
            const fs = require('fs');
            const body = fs.readFileSync('description.md', 'utf8');
            github.rest.pulls.update({
              owner: context.repo.owner,
              repo: context.repo.repo,
              pull_number: context.issue.number,
              body: body
            });

Automated Fixes on Push

name: Auto-fix Issues

on:
  push:
    branches: [main]

jobs:
  fix:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Install Claude Code
        run: npm install -g @anthropic-ai/claude-code

      - name: Fix Linting Issues
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
        run: |
          # Run linter, capture issues
          npm run lint 2>&1 | head -100 > lint-issues.txt || true

          # Only run if there are issues
          if [ -s lint-issues.txt ]; then
            claude --dangerously-skip-permissions --print -p "
              Fix these linting issues:
              $(cat lint-issues.txt)
            "
          fi

      - name: Commit Fixes
        run: |
          git config user.name "Claude Bot"
          git config user.email "claude@example.com"
          git add -A
          git diff --staged --quiet || git commit -m "fix: auto-fix linting issues"
          git push

What Works Well in CI

Use Case	Why It Works
PR review	Stateless, single pass
PR descriptions	Summarization task
Commit message generation	Small, focused input
Documentation updates	Isolated changes
Simple migrations	Pattern-based transforms

What Doesn’t Work Well

Use Case	Why It Fails
Complex debugging	Needs iteration
Large refactors	Requires human judgment
Architectural decisions	Needs context/discussion
Anything requiring /compact	Long context management

Cost Control in CI

Automated runs can burn through tokens fast:

# Use Haiku for simple tasks
- run: |
    claude --print --model haiku -p "..."

# Limit agentic turns per run
- run: |
    claude --print --max-turns 5 -p "..."

# Cache results where possible
- uses: actions/cache@v4
  with:
    path: .claude-cache
    key: claude-${{ hashFiles('**/*.py') }}

Caution

Monitor your CI costs. A misconfigured workflow can trigger hundreds of Claude calls. Set budget alerts in your Anthropic account.

Outputting Structured Data

For programmatic consumption:

# Request JSON output
claude --print -p "
  Analyze this code and output JSON:
  {
    \"issues\": [{\"file\": string, \"line\": number, \"severity\": string, \"message\": string}],
    \"summary\": string
  }

  $(cat src/main.py)
"

# Parse in CI
claude --print -p "..." | jq '.issues[] | select(.severity == "high")'

Debugging CI Runs

When Claude fails in CI:

# Capture full output
- run: |
    claude --print -p "..." 2>&1 | tee claude-output.txt

# Upload as artifact
- uses: actions/upload-artifact@v4
  if: failure()
  with:
    name: claude-debug
    path: claude-output.txt