The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides tools to ingest and evaluate traces and datasets from LLM executions, which creates a vulnerability surface for indirect prompt injection. Ingestion points: Untrusted model outputs and user queries are ingested via functions like client.get_spans_dataframe and run_evals in SKILL.md. Boundary markers: The provided code snippets and prompt templates do not show the use of delimiters or specific instructions to mitigate instructions embedded within the processed data. Capability inventory: The skill environment possesses network communication capabilities for trace ingestion and external API calls, as well as database storage access. Sanitization: No input sanitization or filtering of the model-generated content is documented in the provided implementation examples.

phoenix-observability