The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (markdown or text PRDs) to generate structured output that controls an autonomous agent. While this presents an attack surface for indirect prompt injection where a malicious PRD could attempt to influence the agent's behavior, the skill's instructions focus on logical conversion and do not exhibit malicious intent.
Ingestion points: Processes user-provided PRD text and markdown content.
Boundary markers: No explicit delimiters are used to wrap the untrusted PRD content during the conversion process.
Capability inventory: The skill involves writing to the local file system (prd.json, progress.txt, and archive folders) and referencing external tools like the dev-browser skill.
Sanitization: No specific sanitization or filtering logic is provided for the input text.
[COMMAND_EXECUTION]: The skill instructs the agent to perform local file system operations, including creating directories and copying files for archival purposes (archive/YYYY-MM-DD-feature-name/). These operations are restricted to the local project environment and are consistent with the skill's stated purpose of managing agent tasks.

ralph