The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from the codebase and user input to generate documentation, creating a surface for indirect prompt injection.\n- Ingestion points: Data enters the agent context through the $ARGUMENTS variable and file content retrieved from the codebase via the Read, Glob, and Grep tools in SKILL.md.\n- Boundary markers: The instructions lack explicit requirements for using delimiters or 'ignore' warnings when the collected codebase data is processed by the documentation-writing sub-agent.\n- Capability inventory: The skill is granted Write and Edit permissions on the filesystem, enabling it to create or modify files in the .groove/memory/docs/ directory.\n- Sanitization: A constraint is present to sanitize the $ARGUMENTS topic when used as a filename to prevent path traversal (../), although no content-level sanitization is performed on the ingested data.

groove-work-doc