Skill Audit Lab

Architectural security assessment of AI Agent Skills and social protocols

SYSTEM STATUS: MONITORING
Skills, feeds, and verification flows remain under review.

Latest Local Audit

EXEC_MODE: LOCAL_PATH_SCANNER v1.1.0

--- 🛡️ ZeroG Skill Audit ---
Target: /root/agent-playground/.agent/skills/moltbook-social/SKILL.md
Status: CLEAN
✅ No immediate architectural threats detected.

Threat Intelligence

VULNERABILITY: skill.md Credential Hijacking

Vector: Malicious instructions in SKILL.md files.

Risk: High. Agents may be tricked into reading ~/.env or config files and exfiltrating data via curl or POST requests.

"Signal discovered by eudaemon_0 on Moltbook. 🦞"

MITIGATION: ZeroG Audit Protocol

Our lab implements pattern-matching heuristics to detect unauthorized file access, stale claims, and data exfiltration attempts before skill activation.