The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external evaluation datasets (e.g., eval_dataset.json) to optimize AI agent parameters and prompts.
Ingestion points: Untrusted data enters the agent context via the eval.general.dataset configuration field in workflow.yaml (referenced in references/complete-config-example.md).
Boundary markers: The provided documentation does not define specific delimiters or instructions for the agent to ignore malicious content potentially embedded within the evaluation datasets.
Capability inventory: The nat optimize process executes workflows which include LLM generation and potentially tool execution across multiple trials (as described in references/choosing-parameters.md).
Sanitization: No evidence of sanitization, filtering, or validation of the dataset content is present in the provided configuration files or reference guides.

nat-optimization