Skill Security Detection Engine

Problem

Third-party skills can introduce risky behavior in code, dependencies, or prompts. Before installation, it is hard to tell whether a skill may access sensitive environments, make network calls, exceed permissions, or ship hardcoded credentials. This skill turns that review into a repeatable static check, useful before onboarding, updating, or CI gating.

How it works

It runs in a sandboxed read-only environment with only read access, dependency checks, and constrained analysis, avoiding writes and execution. It scans for:
- Code behavior: exec, system, file deletion or overwrite, network access, environment-variable tampering, privilege escalation
- Secrets: API keys, database connection strings, encryption keys, PII
- Prompt risk: malicious construction, system-instruction injection, context pollution
- Dependencies and compliance: vulnerable packages, unauthorized dependencies, cross-border data transfer, privacy leaks, signature/hash integrity

The implementation combines AST-based static pattern matching, regex-based secret detection, dependency analysis, and heuristic behavior review, then applies weighted dimension scoring to produce Low, Medium, High risk levels and confidence. It can output standard reports or JSON results, making it suitable for Skills Hub integration and CI/CD pre-checks. Treat it as static analysis: false positives are possible, large skills may scan slowly, rules need periodic updates, and high-risk findings still require manual review.