The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by reading source code and prototypes from user-specified file paths. Malicious instructions embedded within these files (e.g., in code comments) could attempt to influence the agent's analysis or the resulting documentation content.
Ingestion points: Analyzes files and directories at the <path> provided by the user (SKILL.md, Phase 2).
Boundary markers: The skill does not define specific delimiters for separating code from instructions, but it uses structured templates for output (Phase 5).
Capability inventory: The skill has access to Read, Glob, Grep, Write, Edit, and Bash tools to analyze implementation and write documentation.
Sanitization: The process includes a mandatory human-in-the-loop review step (Phase 6), where the agent must present the draft for approval before performing any Write operations.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform implementation analysis, such as searching for patterns or listing files. The instructions guide the agent to use these tools for informational purposes rather than system-level configuration changes.

reverse-document