The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided data (PRDs or descriptions) to generate instructions for subsequent autonomous execution.
Ingestion points: Step 1 of the Activation Protocol in SKILL.md reads user-provided file paths or text descriptions.
Boundary markers: The skill structures the output into a Markdown template but does not use explicit security delimiters for interpolated content.
Capability inventory: The skill possesses file read/write capabilities and spawns subagents using the Task tool for review purposes.
Sanitization: The skill includes a "Blind Review Protocol" that specifically instructs a subagent to search for ambiguities, assumptions, and potential failures in the generated prompt, which serves as a robust functional validation and safety checkpoint.
[SAFE]: The skill's operations, including reading local project context (CLAUDE.md) and generating implementation plans, are standard for developer productivity tools and do not involve unauthorized network access, credential harvesting, or obfuscation.

ralph-loop-creator