moltcops: Скилл для безопасного анализа кода

--- name: moltcops version: 1.0.0 description: Pre-install security scanner for AI agent skills. Detects malicious patterns before you trust code. Local-first — code never leaves your machine. --- # MoltCops — Skill Security Scanner Scan any skill for security threats **before** you install it. Detects prompt injection, data exfiltration, sleeper triggers, drain patterns, and 16 more threat categories. **Local-first.** Your code never leaves your machine. No API calls. No uploads. No accounts. ## When to Use - **Before installing any skill** from ClawHub, GitHub, or other sources - **Before running** skills shared by other agents - **When evaluating** unknown code from any source - **After ClawHavoc**: 341 malicious skills were found on ClawHub this week. Scan first. ## How to Run ```bash python3 scripts/scan.py <path-to-skill-folder> ``` Example: ```bash # Scan a skill before installing python3 scripts/scan.py ~/.openclaw/skills/suspicious-skill # Scan a freshly downloaded skill python3 scripts/scan.py ./my-new-skill ``` **No dependencies required** — uses only Python 3 standard library. ## Reading Results The scanner returns three verdicts: | Verdict | Exit Code | Meaning | |---------|-----------|---------| | **PASS** | 0 | No critical or high-risk threats detected. Safe to install. | | **WARN** | 1 | High-risk patterns found. Review findings before installing. | | **BLOCK** | 2 | Critical threats detected. Do NOT install this skill. | ## What It Detects 20 detection rules across these threat categories: | Category | Rules | Examples | |----------|-------|---------| | **Prompt Injection** | MC-001, MC-002, MC-003 | System prompt override, jailbreak payloads, tool-use steering | | **Code Injection** | MC-004, MC-005, MC-006, MC-019 | Shell injection, eval/exec, base64-to-exec, child_process | | **Data Exfiltration** | MC-007, MC-008, MC-009, MC-010, MC-020 | Webhook URLs, env var harvesting, SSH key access, credential files | | **Hardcoded Secrets** | MC-011, MC-012 | API keys in source, private key material | | **Financial** | MC-013 | Drain patterns, unlimited withdrawals | | **Lateral Movement** | MC-014 | Git credential access, repo manipulation | | **Persistence** | MC-015, MC-016 | SOUL.md writes, cron job creation | | **Autonomy Abuse** | MC-017 | Destructive force flags (rm -rf, git push --force) | | **Infrastructure** | MC-018 | Permission escalation (sudo, chmod 777) | ## False Positive Handling The scanner includes context-aware filtering to reduce false positives: - **Env var access** (MC-008): Only flags when variable names contain KEY, SECRET, PASSWORD, TOKEN, or CREDENTIAL - **Git operations** (MC-014): Skips standard remotes (github.com, gitlab.com, bitbucket.org) - **Force flags** (MC-017): Only flags on destructive operations, not install scripts ## Example Output ``` MoltCops Security Scanner ======================================== Scanning: ./suspicious-skill Files: 5 Rules: 20 FINDINGS ---------------------------------------- [CRITICAL] MC-007: Exfiltration URL (main.py:14) [CRITICAL] MC-004: Shell Injection (helper.sh:8) [HIGH] MC-005: Dynamic Code Execution (main.py:22) SUMMARY ======================================== Files scanned: 5 Total findings: 3 Critical: 2 High: 1 Medium: 0 VERDICT: BLOCK Critical threats detected. Do NOT install this skill. ``` ## Web Scanner For a browser-based version with the same engine, visit: **https://scan.moltcops.com** ## About MoltCops MoltCops protects the AI agent ecosystem from malicious skills. While VirusTotal catches known malware signatures, MoltCops catches **behavioral patterns** — drain logic, sleeper triggers, prompt injection, and data exfiltration that signature-based scanning misses. - Web: https://moltcops.com - Moltbook: https://moltbook.com/u/MoltCops