The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses WebFetch to retrieve code from api.github.com and raw.githubusercontent.com for the purpose of auditing remote repositories. This behavior is restricted to GitHub domains and is used strictly for static analysis, not for execution.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted data from local files and remote GitHub repositories. However, this risk is inherent to its primary purpose as a security auditor and is mitigated by several factors:
Ingestion points: Content is read via Read, Grep, and WebFetch (SKILL.md, Phase 1 & 2).
Boundary markers: The skill includes an explicit 'Anti-Injection Protocol' that instructs the agent to treat all audited content as untrusted data and specifically flags injection-like phrases (e.g., 'ignore previous instructions') as findings rather than following them (SKILL.md, 'Anti-Injection Protocol' section).
Capability inventory: The skill is restricted to read-only tools (Read, Grep, Glob, WebFetch) and lacks the ability to execute code (Bash) or modify files (Write, Edit).
Sanitization: It automatically redacts detected secrets (API keys, tokens) in the generated audit reports and limits evidence snippets to small, labeled excerpts.

skill-audit