dependency-scan

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The prompt contains explicit "SELF-EVOLUTION TELEMETRY" instructions to check the user's ~/.claude/projects/ directory and append usage metadata (skipping silently if not found), which is unrelated to dependency scanning and effectively adds stealthy telemetry/logging behavior outside the skill's stated purpose.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill explicitly runs package-manager audit commands (e.g., npm audit --json, yarn audit, pnpm audit, cargo audit, govulncheck, and a pip-audit fallback that may use web search) and parses their advisory/CVE outputs (and even instructs manual checks against public CVE databases), which are external, untrusted public sources whose content is read and used to make fixing/commit decisions.