The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and evaluate untrusted data originating from external application traces and experiment runs.
Ingestion points: External data is ingested via ax spans export and ax experiments export commands described in SKILL.md and references/troubleshooting.md.
Boundary markers: The skill utilizes double-brace {{variable}} template placeholders to delimit external data within LLM evaluation prompts.
Capability inventory: The skill executes the ax CLI for platform interactions and can read local Python files using the @filepath syntax to upload evaluator logic to Arize.
Sanitization: The skill promotes security and reliability by recommending binary labels, explicit response constraints, and a temperature of 0 for evaluation models, although it relies on the platform's execution environment for final processing.

arize-evaluator