# Agent Skill Audit — Agent API

Pre-flight safety check for AI agents. Before you run code from the internet, know whether it's safe.

## Quick Start

```bash
# Get an API key (100 free credits)
curl -X POST https://www.agentskillaudit.ai/waitlist -d "email=you@domain.com"

# Check if a repo is known-safe (free)
curl "https://www.agentskillaudit.ai/api/v1/check?source=https://github.com/owner/repo&source_type=url"

# Audit a repo (1 credit, sync)
curl -X POST https://www.agentskillaudit.ai/api/v1/audits \
  -H "Authorization: Bearer sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"source_type":"url","source":"https://github.com/owner/repo","sync":true}'

# Audit raw skill text
curl -X POST https://www.agentskillaudit.ai/api/v1/audits \
  -H "Authorization: Bearer sk_live_..." \
  -d '{"source_type":"text","source":"skill markdown here...","sync":true}'
```

## Integration

- **[Agent Instructions](/agent/skill.md)** — drop-in system prompt for your agent
- **[MCP Server](/agent/mcp.md)** — native MCP tools, no HTTP code needed

## Response

Sync audits return immediately:
```json
{
  "policy": {"decision": "allow"|"warn"|"block"|"human_review", "reasons": [...]},
  "findings": [{"category": "INJECTION", "severity": "critical", "message": "...", "evidence": "..."}],
  "summary": {"grade": "A"|"B"|"C"|"D"|"F", "score": 0.95, "risk": "low"|"medium"|"high"|"critical"}
}
```

---

## Recent Audits