The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided product descriptions and question responses to generate documentation. This data is interpolated into a PRD template and saved to the filesystem without sanitization, which is a standard behavior for this type of skill but presents an indirect injection surface.
Ingestion points: Initial arguments ($ARGUMENTS) and interactive responses from the AskUserQuestion tool in SKILL.md (Phase 1).
Boundary markers: The PRD template in assets/prd-template.txt does not use specific delimiters or warnings to isolate user-provided content.
Capability inventory: The skill uses Write and Edit tools to save files to the docs/ directory and manages workflow state using TaskCreate and TaskUpdate.
Sanitization: User input is written directly to the output file without escaping or validation.
[COMMAND_EXECUTION]: The skill contains logic in Phase 3 to execute the open system command to preview the generated PRD. The file path includes a user-derived string ([feature-name]). Although the required shell execution tool is not present in the allowed-tools list, this pattern is noted as a potential (though low-risk) command injection surface if the environment allowed execution.

prd