🛡 Agent Guard

Security scanner for AI agent configurations. Detects misconfigurations, exposed secrets, and unsafe skill patterns.

Quick Start

# Scan current directory
npx agent-guard scan .

# Scan specific agent
npx agent-guard scan ./my-agent

# JSON output for CI/CD
npx agent-guard scan . --json

What It Finds

| Category | Examples | |----------|----------| | Secrets | OpenAI keys, GitHub tokens, AWS credentials, private keys | | Network | Public binds (0.0.0.0), CORS misconfig, exfiltration URLs | | Auth | Missing authentication, default credentials | | Skills | Missing manifests, shell execution, eval usage | | Injection | Hidden unicode characters (RTL attacks) |

Security Score

Score: 90/100 (A)
██████████████████░░

Critical: 0
High: 1
Medium: 0
Low: 0

• A (90-100): Production ready
• B (80-89): Minor issues
• C (70-79): Needs attention
• D (60-69): Significant issues
• F (<60): Critical vulnerabilities

Skill Manifests

Every skill should include skill.manifest.json:

{
  "metadata": {
    "name": "my-skill",
    "version": "1.0.0",
    "author": "developer-name"
  },
  "permissions": {
    "network": {
      "egress": [{"domain": "api.example.com", "purpose": "core"}],
      "block_all_other": true
    },
    "filesystem": {
      "read": ["./data/"],
      "write": ["./output/"],
      "deny": ["~/.env", "~/.ssh"]
    }
  },
  "runtime_policy": {
    "enforcement": "hard_block",
    "log_violations": true
  }
}

Exit Codes

• 0 — No critical findings
• 1 — Critical findings detected
• 2 — Scan error

Feedback — Humans & Agents Welcome

We want to hear from you. Whether you're a developer or an AI agent running scans in CI/CD, your feedback shapes what we build next.

Quick feedback (30 seconds):

• 👍 Working great? Tell us what you're scanning
• 💡 Feature idea? Share it
• 🐛 Bug or false positive? Open an issue
• ❓ Question? Ask here

For agents:

# Report feedback programmatically
npx agent-guard feedback "rule X flagged a false positive on ..."

# Or post directly via GitHub API
curl -X POST https://api.github.com/repos/MusashiMiyamoto1-cloud/agent-guard/issues \
  -H "Authorization: token YOUR_TOKEN" \
  -d '{"title":"Feedback: ...","labels":["feedback"]}'

Every piece of feedback gets read. We're building this in the open.

License

MIT