The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes multiple external AI CLI tools (e.g., claude, gemini, codex, opencode) as subprocesses. This is the primary intended function and uses fixed templates for command construction.
[EXTERNAL_DOWNLOADS]: Fetches and executes ACP adapters using npx for Claude and Codex. These are sourced from well-known repositories, including the trusted @anthropic-ai organization.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface (Category 8). Untrusted data enters via the user-provided question and ingested file content (SKILL.md). No explicit boundary markers or instructions to ignore embedded commands are included in the command templates. The skill's capabilities include executing shell-based tools. Sanitization is implemented via automated credential redaction in tool responses.
[SAFE]: Implements strong security mitigations including automated redaction of sensitive API keys and tokens from tool responses and robust path validation for the --context flag to prevent directory traversal by verifying canonical paths remain within the workspace.

consult