hiro-agent

AI security review agent for code, plans, and infrastructure.

What this project does

• Reviews code diffs for security issues
• Reviews implementation plans with threat-modeling prompts
• Reviews infrastructure configs for security misconfigurations
• Scans repositories with multi-skill, wave-based security investigations (experimental)

It integrates with Claude Code, Cursor, VSCode Copilot, and Codex CLI to enforce review workflows before commits and plan finalization.

Install

# Recommended (isolated environment, works on macOS/Linux)
pipx install hiro-agent

# Or with pip in a virtual environment
pip install hiro-agent

Quick Start

# Set up hooks for your AI coding tools
hiro setup

# Review code changes
git diff | hiro review-code

# Review an implementation plan
cat plan.md | hiro review-plan

# Review infrastructure configuration
hiro review-infra main.tf

# Experimental: full-repo scan
hiro scan

Commands

Setup Options

hiro setup                # Auto-detect all tools
hiro setup --claude-code  # Claude Code only
hiro setup --cursor       # Cursor only
hiro setup --vscode       # VSCode Copilot only
hiro setup --codex        # Codex CLI only

Configuration

Set HIRO_API_KEY to connect to the Hiro platform for organizational context (security policies, memories, org profile). Without it, reviews still run using your ANTHROPIC_API_KEY directly.

export HIRO_API_KEY=hiro_ak_...     # Optional: Hiro platform context
export ANTHROPIC_API_KEY=sk-ant-... # Required if HIRO_API_KEY not set

More details: docs/configuration.md

How It Works

1. hiro setup installs hook scripts in .hiro/hooks/ and configures your AI coding tool to call them
2. Hooks track file modifications and block commits until hiro review-code has run
3. Hooks track plan creation and block finalization until hiro review-plan has run
4. Review agents use claude-agent-sdk to spawn a Claude instance that performs the security review
5. When connected to Hiro (HIRO_API_KEY), reviews are enriched with your org's security policy, accepted risks, and architecture context

Stability

• Primary/release-gated workflows: review-code, review-plan
• Secondary workflow: review-infra
• Experimental workflow: scan

Architecture and Methodology

• Architecture: docs/architecture.md
• Scan methodology: docs/scan-methodology.md
• Troubleshooting: docs/troubleshooting.md
• Development: docs/development.md

Contributing

• Contribution guide: CONTRIBUTING.md
• Code of conduct: CODE_OF_CONDUCT.md
• Security reporting: SECURITY.md

License

MIT