The Agent Skills Directory

[COMMAND_EXECUTION]: The skill guides the user to execute Python scripts using the uv tool. This is a standard pattern for executing localized logic within an AI agent environment and is used here to interact with the SAP AI Core API.
[EXTERNAL_DOWNLOADS]: The download_evaluation_results.py script uses the boto3 library to download evaluation artifacts (specifically a SQLite database) from a user-specified S3 bucket. This behavior is documented, intended for result analysis, and uses standard AWS credential chains.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted dataset files (CSV, JSON, JSONL) for benchmarking LLMs. This presents a common attack surface for indirect prompt injection where malicious instructions could be embedded in the test data.
Ingestion points: scripts/create_evaluation_configuration.py reads dataset files from object storage linked via the --test-dataset-path parameter.
Boundary markers: The prompt templates located in the templates/ directory utilize standard variable interpolation markers (e.g., {{?question}}) to delimit data inputs.
Capability inventory: The skill has the capability to initiate executions in SAP AI Core and perform read-only network operations to S3 buckets.
Sanitization: The skill relies on underlying API validation and standard parsing libraries; it does not implement manual sanitization of the dataset content before prompt interpolation.

genai-hub-prompt-evaluation