Claude Code Security Reviewer Documentation

To use the Claude Code Security Reviewer in your repository, create a GitHub Actions workflow file (e.g., .github/workflows/security.yml). The action requires pull-requests: write permissions to post comments on your PRs.

Ensure you have a CLAUDE_API_KEY stored in your GitHub repository secrets. This key must be enabled for both the Claude API and Claude Code usage.

name: Security Review

permissions:
  pull-requests: write  # Needed for leaving PR comments
  contents: read

on:
  pull_request:

jobs:
  security:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          ref: ${{ github.event.pull_request.head.sha || github.sha }}
          fetch-depth: 2
      
      - uses: anthropics/claude-code-security-review@main
        with:
          comment-pr: true
          claude-api-key: ${{ secrets.CLAUDE_API_KEY }}

You can extend the Claude Code Security Reviewer's audit by adding organization-specific vulnerability categories. This allows the tool to check for concerns related to your specific technology stack (e.g., GraphQL, gRPC), compliance requirements (e.g., GDPR, HIPAA), or industry-specific patterns.

To use custom instructions, create a text file containing your categories and reference it in your workflow using the custom-security-scan-instructions input.

- uses: anthropics/claude-code-security-review@main
  with:
    custom-security-scan-instructions: .github/custom-security-categories.txt

You can tailor the security analysis to your specific environment by providing custom filtering instructions via the false-positive-filtering-instructions input. This allows you to define specific exclusions, quality criteria, and precedents to reduce noise in security findings.

To use custom instructions:

1. Create a plain text file containing your instructions (e.g., .github/false-positive-filtering.txt).
2. Reference the file path in your workflow configuration.

- uses: anthropics/claude-code-security-review@main
  with:
    false-positive-filtering-instructions: .github/false-positive-filtering.txt

To customize the behavior of the /security-review command (e.g., adding organization-specific false positive filtering instructions), follow these steps:

1. Copy the security-review.md file from this repository to your project's .claude/commands/ folder.
2. Edit the security-review.md file to include your custom instructions.

To use the SAST Evaluation Tool, ensure your environment meets the following requirements:

• Python: version 3.9 or higher
• Git: version 2.20 or higher (required for git worktree support)
• GitHub CLI: gh must be installed for API access

Required Environment Variables:

• ANTHROPIC_API_KEY: Necessary for accessing the Claude API.

Recommended Environment Variables:

• GITHUB_TOKEN: Recommended to avoid GitHub API rate limits.

When implementing custom filtering instructions, follow these best practices to maintain high-quality security analysis:

• Start with defaults: Use the action's default instructions as a baseline and modify them only as you encounter specific false positives.
• Be specific: Include details about your unique security architecture (e.g., "We use AWS Cognito for all authentication").
• Document assumptions: Explain the reasoning behind exclusions (e.g., "k8s resource limits prevent DOS").
• Version control: Keep your filtering instructions in your repository to track changes alongside your code.
• Team review: Ensure your security team reviews and approves any changes to the filtering logic.

You can evaluate the Claude Code Security Reviewer on a specific GitHub PR by running the claudecode.evals.run_eval module. You must provide the PR in the format owner/repo#pr_number.

python -m claudecode.evals.run_eval example/repo#123 --verbose

Prompt Injection Warning

This action is not hardened against prompt injection attacks. It should only be used to review trusted pull requests.

Recommended Mitigation

To ensure workflows only run after a maintainer has reviewed the PR, configure your repository to use the "Require approval for all external contributors" option in GitHub settings.

The action provides the following outputs: