Plugin Approval Guard

"VirusTotal for OpenClaw skills" — static analysis trust scanner.

Commands

Run via: python3 scripts/plugin-approval-guard.py [options]

Commands

• scan — Scan a skill directory for security issues
• batch — Scan all installed skills
• report — Generate trust report (text/json/markdown)
• allow — Add a skill to the local trust allowlist
• deny — Block a skill from execution
• list — Show trust status of all installed skills

Options

• --skill PATH — Path to skill directory to scan
• --workspace PATH — Workspace path (default: auto-detect)
• --format text|json|markdown — Output format (default: text)
• --severity low|medium|high|critical — Minimum severity to report
• --output PATH — Save report to file
• --verbose — Show code snippets with findings

Trust Score Categories

• Code Safety (40%) — No eval/exec, no shell injection, no obfuscation
• Network Access (25%) — Declared vs hidden network calls, data destinations
• File Access (20%) — Scope of filesystem access, no writes outside workspace
• Permission Scope (15%) — Minimal permission principle, no escalation patterns

Detection Patterns

See references/threat-patterns.md for the full list of malicious patterns detected.

Integration

Leverages OpenClaw 3.28 plugin approval hooks. When a scanned skill has trust score below C, actions from that skill trigger approval prompts.