The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from user-provided draft files.
Ingestion points: Reads draft content from a user-specified path (<draft-path>) in Step 1 of the workflow.
Capability inventory: The skill is restricted to Read, Glob, and Grep tools and explicitly forbids writing any files or making network requests.
Boundary markers: While it uses a structured template for the sub-agent task, it lacks explicit XML-style delimiters to separate the untrusted draft content from the scoring instructions.
Sanitization: Includes a mandatory grep check to identify and remove keywords (e.g., performance metrics or historical data) that could bias the scoring sub-agent.

cheat-score