The Agent Skills Directory

[SAFE]: The skill is a standard product management tool with no indicators of malicious activity, data exfiltration, or obfuscation.
[COMMAND_EXECUTION]: The skill performs file system writes and git commits to save the PRD in the docs/plans/ directory. These operations are restricted to the intended output path and are part of the core functionality.
[PROMPT_INJECTION]: The skill's workflow involves reading existing codebase files and documentation, which creates an indirect prompt injection surface.
Ingestion points: Local codebase, documentation, and tickets.
Boundary markers: Absent; the agent is not explicitly told to ignore instructions within these files.
Capability inventory: File creation/writing in docs/plans/ and version control commit operations.
Sanitization: Absent; content from the codebase is processed directly to generate dialogue questions. The risk is considered safe because of the interactive nature of the skill and the human-in-the-loop requirement for the final document.

product-requirements