The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute grep and other shell commands to inspect target files for destructive patterns. This is part of its core functionality as a static analysis auditor.
[EXTERNAL_DOWNLOADS]: The skill requires the PyYAML Python library and executes a local script scripts/preflight.sh to verify the environment. These are standard dependencies for its stated purpose.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it ingests and processes untrusted SKILL.md files from a user-provided path.
Ingestion points: Target files are read in Step 1 and their content is passed to an LLM helper in Step 2.
Boundary markers: Step 2 uses a structured prompt to guide the LLM to output only JSON, which provides some constraint on the output format.
Capability inventory: The skill has access to Read and Bash tools, and can write report files to the filesystem.
Sanitization: There is no evidence of content sanitization or filtering of the target files before they are analyzed by the LLM or passed to shell commands.

audit