SkillSpector
A local security scanner checks AI agent skills for risky instructions and code, with optional LLM analysis and Apache 2.0 licensing.
SkillSpector scans AI agent skills before installation or publication. It helps people reviewing Claude Code, Codex CLI, Gemini CLI and MCP skills spot risks in their instructions, permissions and code.
- Security checks: Look for prompt injection, data exfiltration, excessive permissions, dangerous code, memory poisoning and MCP tool misuse without executing the skill.
- Flexible input: Review Git repositories, URLs, zip archives, local directories or individual files.
- Dependency checks: Query OSV.dev for known vulnerabilities in declared dependencies, with a bundled fallback when the service is unreachable.
- Optional LLM review: Compare a skill's stated purpose with its apparent behavior to flag misleading descriptions, unclear triggers and subtle policy issues.
- Review reports: Get risk scores, severity labels and recommendations in terminal, Markdown, JSON or SARIF output, and suppress accepted findings with baselines.
Static analysis runs locally and keeps file contents on your machine. Dependency lookups still send package names and versions to OSV.dev, without requiring an API key. Optional LLM analysis sends eligible file contents to the configured provider; OpenAI is a supported choice. Model review doesn't remove deterministic findings from reports.
SkillSpector is open source under Apache 2.0 and also provides SkillEvaluator's security checks. It can produce false positives and doesn't isolate or contain a skill after installation.
