The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted input data (the document string) by interpolating it directly into LLM prompts without explicit boundary markers or instructions to disregard embedded commands.
Ingestion points: The ensemble_review and single_review functions in SKILL.md accept a document string which is user-controlled data.
Boundary markers: The prompt template in SKILL.md uses simple newline separation: "Review this document for compliance issues:\n\n{document}". It lacks robust delimiters (e.g., XML tags or block markers) to isolate the data from the instructions.
Capability inventory: The skill uses the anthropic client to execute LLM calls. While it doesn't directly perform shell or file operations, it is designed to evaluate code and compliance documents.
Sanitization: No sanitization or escaping of the input {document} is performed before it is sent to the LLM.

ensemble-review