The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The retrieveAndGenerate function in SKILL.md demonstrates a RAG (Retrieval-Augmented Generation) pattern that interpolates untrusted data from an external vector database and a user query into the model prompt.
Ingestion points: The context (retrieved from vectorDb) and query (user input) variables in SKILL.md.
Boundary markers: The prompt construction uses simple text labels like Context:\n${context} but lacks robust delimiters or explicit instructions to the model to ignore potential instructions embedded within the retrieved context.
Capability inventory: The agentLoop in SKILL.md implements function calling (executeToolCall), which could be misdirected if an injection occurs via the prompt surface.
Sanitization: The provided code snippets do not implement sanitization, filtering, or escaping for external content before it is interpolated into the prompt template, although the 'Anti-Patterns' section correctly identifies a lack of validation as a risk.

llm-integration