Claude API Setup Guide for GitHub CI/CD

TL;DR: This enables GitHub Actions to automatically run the Agent Auditor (.claude/agents/agent-auditor.md) on a schedule, auditing all your AI agents for quality issues without manual intervention.

This guide covers how to configure the Claude API so GitHub Actions CI/CD can automatically run the Agent Auditor agent to perform weekly quality audits of your AI agent library.

What This Enables: Automated CI/CD Agent Auditing

The API setup allows GitHub Actions to run the Agent Auditor automatically:

The GitHub Actions Workflow

• 🔄 Runs automatically every Sunday at midnight UTC
• 🤖 No manual intervention - CI/CD handles everything
• 📊 Creates audit reports directly in your repository
• ✅ Can block merges if agents fail quality checks
• 🔔 Can notify on Slack/email when issues are found

What the Agent Auditor Checks

The .claude/agents/agent-auditor.md agent that GitHub Actions runs will:

• 📋 Audit all agents in .claude/agents/ for quality and completeness
• 🔍 Validate frontmatter ensuring proper agent-type and allowed-tools
• 📝 Check documentation for clear instructions and success criteria
• 🎯 Verify task focus ensuring each agent has a single, clear purpose
• 🚨 Flag security issues like overly broad tool access

Why Use GitHub Actions for This?

Without automated CI/CD auditing:

• ❌ You manually review each agent (time-consuming)
• ❌ Quality issues slip through to production
• ❌ No consistent quality enforcement across team members
• ❌ No audit trail of agent quality over time

With GitHub Actions + Claude API:

• ✅ Automatic weekly quality audits
• ✅ Consistent quality standards enforced
• ✅ Historical reports tracked in git
• ✅ Team notified of issues automatically

Quick Setup: Enable GitHub Actions CI/CD for Agent Auditing

1. Get Your Claude API Key

• Visit Anthropic Console
• Navigate to API Keys section
• Create a new key for this project

2. Add API Key to GitHub Secrets (Required for CI/CD)

1. Go to your repository on GitHub
2. Navigate to Settings → Secrets and variables → Actions
3. Click "New repository secret"
4. Add the secret:
   • Name: ANTHROPIC_API_KEY
   • Value: Your API key (sk-ant-...)
5. Click "Add secret"

⚠️ NEVER commit your API key to the repository!

3. GitHub Actions Workflow is Now Enabled

Once the secret is added, the .github/workflows/agent-audit.yml workflow will:

• Run automatically: Every Sunday at midnight UTC
• Run manually: Via Actions tab → Agent Audit → Run workflow
• Use the API key: To run Claude and audit your agents
• Commit reports: Back to your repository for review

Local Development Setup

For local API usage with agents:

Option 1: Environment Variable

export ANTHROPIC_API_KEY="sk-ant-..."
# Now run your agent commands

Option 2: .env File (Git-Ignored)

echo "ANTHROPIC_API_KEY=sk-ant-..." > .env
# The .env file is already in .gitignore

Option 3: Claude Code CLI Config

claude config set api-key sk-ant-...

Security Best Practices

DO ✅

• Store API keys in environment variables or secrets
• Use different keys for development and production
• Rotate keys regularly
• Limit key permissions when possible
• Monitor usage via the Anthropic Console

DON'T ❌

• Commit API keys to version control
• Share keys between team members
• Log API keys in console output
• Include keys in error messages
• Use production keys for testing

Cost Considerations

API Pricing

• Claude 3.5 Sonnet: ~$3 per million input tokens
• Typical agent audit: ~10,000 tokens (~$0.03)
• Weekly audits: ~$1.50/month

Optimization Tips

• Use caching for repeated analyses
• Batch related operations
• Filter inputs to relevant content only
• Monitor usage in Anthropic Console

Workflow Examples

Automated Agent Audit (Primary Use Case)

# .github/workflows/agent-audit.yml
# Runs the Agent Auditor to check all agents in your library
- uses: rmurphey/claude-code-cli@v1
  with:
    api-key: ${{ secrets.ANTHROPIC_API_KEY }}
    command: run .claude/agents/agent-auditor.md

This runs the Agent Auditor which:

1. Scans all .claude/agents/*.md files
2. Validates each agent's configuration
3. Creates a report at .claude/agents/reports/agent-audit-[date].md

Manual Local Audit

# Run the Agent Auditor locally (requires API key configured)
claude run .claude/agents/agent-auditor.md

# View the generated report
cat .claude/agents/reports/agent-audit-*.md

Other Available Agents (Also Require API)

# Documentation Auditor - checks all documentation
claude run .claude/agents/documentation-auditor.md

# Session Insights Analyzer - analyzes development patterns
claude run .claude/agents/session-insights-analyzer.md

# Command Analyzer - audits command templates
claude run .claude/agents/command-analyzer.md

Troubleshooting

Common Issues

"Invalid API Key"

• Verify key starts with sk-ant-
• Check for extra spaces or quotes
• Ensure key hasn't been revoked

"Rate Limit Exceeded"

• Default: 50 requests per minute
• Solution: Add delays between requests
• Consider upgrading your plan

"Workflow Not Running"

• Check Actions are enabled in repository settings
• Verify secret name is exactly ANTHROPIC_API_KEY
• Check workflow file syntax

"Permission Denied"

• Ensure GitHub Actions has write permissions
• Check branch protection rules
• Verify workflow permissions in settings

Debug Mode

Enable verbose logging:

DEBUG=claude* npm run agent:audit

Advanced Configuration

Custom Model Selection

// In your agent configuration
const config = {
  model: 'claude-3-5-sonnet-20241022',
  maxTokens: 4096,
  temperature: 0.7
};

Webhook Notifications

Configure GitHub to notify on audit completion:

1. Settings → Webhooks → Add webhook
2. Payload URL: Your notification endpoint
3. Events: Workflow runs

Multi-Environment Setup

# Different keys per environment
production:
  secret: ANTHROPIC_API_KEY_PROD
staging:
  secret: ANTHROPIC_API_KEY_STAGING
development:
  secret: ANTHROPIC_API_KEY_DEV

Resources

• Anthropic API Documentation
• GitHub Secrets Documentation
• Claude Code CLI
• Pricing Calculator

Support

• Issues: GitHub Issues
• Community: Discussions
• API Support: Anthropic Support