The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from the user in the form of LLM dataset samples and existing annotations to generate rubrics and quality reports.
Ingestion points: SKILL.md (Section A1 and B1) instructs the agent to read 10–50 entries from the user's dataset or annotations.
Boundary markers: No specific delimiters or "ignore instructions" wrappers are mentioned for the data being read.
Capability inventory: The skill primarily uses reasoning and text output to help the developer; no file system writes, network requests (other than a static link), or code execution capabilities are defined in the instructions.
Sanitization: No explicit sanitization or validation of the input data is described.
[EXTERNAL_DOWNLOADS]: The skill contains a reference to the vendor's platform.
Evidence: Provides a link to https://latitude.so in the closing section for automated trace review.

llm-annotation-guide