Skip to content

AGENTS_CONFIG.md

Latest commit

 

History

History
165 lines (118 loc) · 3.79 KB

AGENTS_CONFIG.md

File metadata and controls

165 lines (118 loc) · 3.79 KB

Code Review Agents Configuration

This document explains how to configure AI-powered code review agents for Code Review Guardian.

Overview

Code Review Guardian supports AI-powered code review agents that can automatically review your pull requests and merge requests. These agents analyze your code for quality, security, and best practices.

Configuration

Environment Variables

Add the following environment variables to your .env file:

# Git Provider API Token
# Required for posting comments and reviews to PRs/MRs
GIT_TOKEN=your_github_or_gitlab_token_here

# Optional: AI Agent Configuration
# AI_AGENT_PROVIDER=openai|anthropic|github_copilot
# AI_AGENT_MODEL=gpt-4|claude-3|etc
# AI_AGENT_TEMPERATURE=0.7

Getting Your Git Token

For detailed step-by-step instructions on creating accounts and obtaining tokens, see TOKEN_SETUP.md.

The quick reference for environment variable format:

# GitHub
GIT_TOKEN=ghp_your_github_token_here

# GitLab
GIT_TOKEN=glpat-your_gitlab_token_here

# Bitbucket
GIT_TOKEN=your_bitbucket_app_password_here

Agent Providers

Code Review Guardian supports multiple AI agent providers:

GitHub Copilot

GitHub Copilot can be used for code review when running in GitHub Actions:

# code-review-guardian.yaml
agents:
  provider: github_copilot
  enabled: true

OpenAI

For OpenAI models (GPT-4, GPT-3.5, etc.):

# code-review-guardian.yaml
agents:
  provider: openai
  model: gpt-4
  enabled: true

Required environment variable:

OPENAI_API_KEY=your_openai_api_key_here

Anthropic Claude

For Anthropic Claude models:

# code-review-guardian.yaml
agents:
  provider: anthropic
  model: claude-3-opus
  enabled: true

Required environment variable:

ANTHROPIC_API_KEY=your_anthropic_api_key_here

Agent Configuration

Configure agents in your code-review-guardian.yaml:

agents:
  enabled: true
  provider: openai  # openai, anthropic, github_copilot
  model: gpt-4
  temperature: 0.7
  
  # What the agent should review
  review_scope:
    - code_quality
    - security
    - performance
    - best_practices
    - documentation
  
  # Agent behavior
  behavior:
    suggest_fixes: true
    explain_issues: true
    provide_examples: true
    severity_threshold: medium  # low, medium, high, critical

Usage

Once configured, the agents will automatically review your code when you run:

./code-review-guardian.sh --agent-review

Or integrate into your CI/CD pipeline:

# .github/workflows/code-review.yml
- name: Run Code Review Guardian with AI Agents
  env:
    GIT_TOKEN: ${{ secrets.GIT_TOKEN }}
    OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
  run: |
    ./code-review-guardian.sh --agent-review --post-comment

Best Practices

  1. Keep tokens secure: Never commit tokens to your repository. Use environment variables or secrets.
  2. Review agent suggestions: AI agents provide suggestions, but always review them before applying.
  3. Configure severity threshold: Set appropriate severity levels to avoid too many false positives.
  4. Test locally first: Test agent reviews locally before enabling in CI/CD.

Troubleshooting

Agent not working

  • Check that GIT_TOKEN is set correctly
  • Verify API keys are valid (for OpenAI/Anthropic)
  • Check agent provider is enabled in configuration

Too many suggestions

  • Increase severity_threshold in agent configuration
  • Reduce temperature for more focused reviews
  • Adjust review_scope to focus on specific areas

Additional Resources