The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in multiple code examples.
Ingestion points: Untrusted data enters the agent context via the text variable in llm_summarize, the document variable in query_focused_summarize, and the source variable in check_factual_consistency (all within SKILL.md).
Boundary markers: No explicit delimiters (such as XML tags or unique markers) or instructions to ignore embedded commands are used when interpolating these variables into the prompt strings.
Capability inventory: The skill uses standard NLP libraries and LLM completion calls; no dangerous file system or shell execution tools are enabled in the snippets.
Sanitization: No sanitization or validation of the input text is performed before it is used to construct LLM prompts.

Summarization Engine