The Agent Skills Directory

[SAFE]: The skill is a specialized security auditor for AI agent extensions. All identified sensitive patterns (e.g., credential paths like .ssh/id_rsa, exfiltration endpoints, and prompt injection markers) are part of its detection logic or documentation and do not represent functional malicious behavior within the skill itself.
[SAFE]: Shell command usage is strictly limited via allowed-tools to the skillspector security scanner with the --no-llm flag, preventing data exfiltration to external LLM providers during automated scans and restricting the execution environment.
[SAFE]: The skill includes robust mitigation strategies for handling sensitive data, including mandatory redaction of secrets to prefix + last 4 characters and a strict read-only policy for audited files.
[INDIRECT_PROMPT_INJECTION]: As an extension auditor, the skill processes untrusted data (third-party extension code). It mitigates this risk by providing specific instructions to the agent to treat findings as data for analysis, avoid execution of audited scripts, and perform all dynamic validation in user-controlled sandboxes.
[SAFE]: The included utility scripts (to-sarif.mjs, validate-findings.mjs) are local Node.js tools for processing audit results. They use standard library modules (fs, url) to read and validate JSON findings and do not contain any obfuscated code or remote fetch operations.

agent-supply-chain-audit