openclaw-security
Description
This skill enforces the SEP-2026 Security Protocol. It provides tools to audit source code for malicious patterns, obfuscation, and unauthorized network capability. It is the Gatekeeper; no external code (Skills, MCP Servers, Python scripts) should be integrated without passing this audit.
Usage
Use this skill whenever you need to:
- Audit a new Python file, Skill, or MCP Server before using or installing it.
- Verify the safety of a downloaded script.
- Check for supply chain attacks (typosquatting).
Commands
1. Deep Audit (Source Code)
Run the Shield Pro 2.0 static analyzer on a target directory or file.
python3 skills/openclawSecurity/scripts/shield_pro.py --target <path_to_scan> --format json
Output: A JSON report containing CRITICAL, HIGH, and MEDIUM findings.
Action:
- If
CRITICALorHIGH> 0: REJECT the code. - If
MEDIUM> 0: HALT and ask User for manual review.
2. Supply Chain Check
Check requirements.txt or package.json for typosquatting and version pinning issues.
python3 skills/openclawSecurity/scripts/shield_pro.py --target <path_to_requirements> --mode supply-chain
3. Layer 2: External Defense Arsenal (Recommended)
If available in the environment, the Agent MUST run these SOTA tools to augment Shield Pro's findings:
| Tool | Purpose | Command |
|---|---|---|
| Bandit | Python Static Analysis (Common Vulnerabilities) | bandit -r <target_dir> -f json |
| Safety | Python Dependency Check (Known CVEs) | safety check -r requirements.txt --json |
| NPM Audit | Node.js Dependency Check | npm audit --json (in package dir) |
| Trivy | Filesystem/Container Scan (Secrets/Vulns) | trivy fs <target_dir> --format json |
| Garak | LLM/Prompt Injection Testing | garak --model_type <type> --model_name <name> |
Note: If a tool is missing, log a warning in the final report but proceed with
shield_profindings.
Workflow: The SEP-2026 Gatekeeper Process
- Acquire: Download or locate the target code.
- Isolate: ensure the code is not executed.
- Scan: Run
shield_pro.pyon the target. - Evaluate:
- Pass: No Critical/High issues.
- Fail: Any Critical/High issue found. Reporting potential Line of Code (LOC) and threat type.
- Report: Present the findings to the user.
Principles
- Zero Trust: treat all input as malicious.
- No Execution: Never run the target code during the audit.
- Paranoia: False positives are better than missed malware.