The Agent Skills Directory

[SAFE]: The skill implements a structured workflow for auditing project documentation (CLAUDE.md, AGENTS.md) without executing untrusted code or downloading external payloads.
[SAFE]: Bash tool usage is strictly limited via a whitelist to benign, read-only commands (e.g., ls, cat, grep, md5sum, git ls-files) used exclusively for file discovery and path verification.
[SAFE]: Data privacy is maintained during environment variable auditing (Phase 3f); the instructions explicitly command the agent to identify required variable names but never to reveal or report their actual values.
[SAFE]: The workflow enforces a mandatory human-in-the-loop requirement, using AskUserQuestion to obtain approval for every proposed change, and specifically forbids autonomous commits or deletions.
[SAFE]: Comprehensive analysis of the provided research and reference documents shows no evidence of prompt injection, obfuscated URLs, or hidden malicious logic. All external references point to well-known technology organizations and research papers.

context-audit