The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to process untrusted external content (transcripts, documents, support threads), which is a common vector for indirect prompt injection attacks.
Ingestion points: SKILL.md (Steps 2, 3, 5, 6) and examples.md (Examples 1, 2, 3) define inputs for text, transcript, thread, and document which are filled with untrusted user or external data.
Boundary markers: The provided dspy.Signature definitions do not include delimiters or instructions for the model to ignore potential commands embedded in the source text.
Capability inventory: The skill only demonstrates LLM inference patterns. It does not include subprocess execution, file system modifications, or network operations beyond the standard model provider calls.
Sanitization: No input validation or output sanitization logic is implemented in the tutorial snippets.

ai-summarizing