ClawShield

ClawShield detects high-risk misconfigurations in OpenClaw agents before they become exposed attack surfaces.

It is the first release under the PolicyGate umbrella — a runtime policy enforcement framework for AI agents.

Why This Exists

AI agents are often deployed:

• Publicly bound to 0.0.0.0
• With authentication disabled
• Inside privileged or root containers
• With API keys sitting in .env files
• Without file permission hardening

These are not theoretical risks — they are common misconfigurations.

ClawShield surfaces them deterministically and exits non-zero in CI when thresholds are exceeded.

What ClawShield Checks

Network Exposure

• Public bind address (0.0.0.0, ::)
• Authentication disabled while publicly exposed

Container Posture

• Containers running as root
• Containers running in privileged mode

Secrets Handling

• API keys present in .env files
• API key references inside config files

File Permissions

• World-writable config files
• World-readable or world-writable .env files

What ClawShield Does NOT Check

• Runtime exploitability
• Kernel vulnerabilities
• Docker daemon hardening
• Firewall configuration
• Intrusion detection
• Secrets entropy analysis
• Cloud IAM posture

ClawShield is a static audit tool, not a runtime protection system.

Quick Start (Users)

pip install clawshield

Run audit:

clawshield path/to/openclaw.yaml

JSON mode:

clawshield --json path/to/openclaw.yaml

Fail CI on severity threshold:

clawshield --fail-on high path/to/openclaw.yaml

Severity ranking:

low < medium < high < critical

Exit Codes

Code	Meaning
0	No findings at or above threshold
1	Findings at or above threshold

Example JSON Output

{
  "meta": {
    "schema_version": "0.1",
    "tool_version": "0.3.0",
    "policy_path": "clawshield/policies/vps_public.yaml"
  },
  "facts": [
    {
      "key": "network.bind_address",
      "value": "0.0.0.0",
      "source": "openclaw_config:openclaw.yaml"
    },
    {
      "key": "runtime.auth_enabled",
      "value": false,
      "source": "openclaw_config:openclaw.yaml"
    }
  ],
  "findings": [
    {
      "rule_id": "NET-001",
      "title": "Public bind address with authentication disabled",
      "severity": "critical",
      "confidence": "high",
      "evidence": [],
      "recommended_actions": ["ACT-ENABLE-AUTH"],
      "autofix_available": true
    }
  ]
}

JSON output is deterministic and schema-versioned. Golden tests lock the schema to prevent drift.

Architecture

ClawShield consists of:

• Scanners — Collect facts from runtime and configuration
• Policy Engine — Evaluates YAML rules against collected facts
• Structured Output — Designed for automation and CI pipelines

Scanners are modular and isolated from the engine core.

Roadmap

• Continuous monitoring mode
• Additional runtime adapters
• Expanded Docker hardening checks
• Policy bundles
• Advanced secrets detection
• Signed policy packs

Status

Early release. Actively evolving.

Feedback and contributions welcome.

License

Apache 2.0

Security Disclaimer

ClawShield surfaces rule-based misconfigurations according to the active policy set. It does not guarantee system security.