Security audit tool for ClawHub skills. Scans a skill directory with 65 detection patterns, anti-obfuscation analysis, and dual rating system (Security + Compliance). v0.6.1 fixes batch scan performance by skipping venv/node_modules directories. Use when: installing a new skill, reviewing skill safety, or auditing permissions.
Security audit tool for ClawHub skills. Scans a skill directory with 65 detection patterns, anti-obfuscation analysis, and dual rating system (Security + Compliance). v0.6.1 fixes batch scan performance by skipping venv/node_modules directories. Use when: installing a new skill, reviewing skill safety, or auditing permissions.
Skill Shield v0.6.1 — Security Auditor
Scan any skill directory for permissions and dangerous patterns. Get a safety rating before you install.
Full rule definitions for all 65 detection patterns
CWE taxonomy references (MITRE CWE 4.14)
Partial fingerprints for deduplication across runs
Security severity scores (2.0-10.0 scale)
Skill-shield metadata in run properties (ratings, recommendation)
v0.4.0 — Security Tool False Positive Fix
String Literal Context Detection
Regex patterns and string constants inside security tools (scanners, auditors) are no longer flagged as dangerous code. The scanner now recognizes when a pattern like rm -rf or curl POST appears inside a string literal (quotes, regex, array) and reduces severity accordingly.
Ignore-Next-Line Support
Add # skill-shield: ignore-next-line above any line to suppress the next finding. Useful for known-safe patterns in security tools.
Results
5 security audit tools reclassified from F to A/C/D (agents-skill-security-audit, ai-skill-scanner, skulk-skill-scanner, aoi-prompt-injection-sentinel, aoi-sandbox-shield-lite)
0 regressions on known-safe skills
v0.3.0 — Dual Rating System
Security Rating (code safety)
Based purely on dangerous code patterns found in executable files. Not affected by permission declarations.
Compliance Rating (documentation quality)
Based on permission declaration completeness: does SKILL.md declare the tools actually used in code?
Recommendation
Combines both ratings into an actionable recommendation:
install — Security A/B + Compliance A/B
install_with_review — Code is safe but permissions undeclared (likely poor docs, not malicious)
review_required — Security patterns flagged
do_not_install — Significant security concerns
documentation_only — No executable code (pure SKILL.md guidance)
False Positive Fixes (v0.3.0)
JS template literals (backticks) no longer flagged as shell execution
Variable names (hostname, whoami) no longer flagged as commands
Shell script normal $VAR usage no longer floods findings
