The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by interpolating untrusted user data directly into LLM prompt templates.
Ingestion points: Functions llm_sentiment_analysis and comparative_sentiment in SKILL.md take a text parameter from external sources.
Boundary markers: The skill uses double-quote delimiters "{text}" to separate input from instructions.
Capability inventory: The skill is designed to return structured JSON output via an LLM completion tool.
Sanitization: No explicit sanitization or escaping of the input string is performed before it is placed into the prompt template.

Sentiment Analyzer