The Agent Skills Directory

[SAFE]: The skill consists of local Python scripts that use the standard library to perform statistical and feasibility calculations. No network requests, external dependencies, or suspicious commands were found. The configuration management follows standard practices by storing user preferences locally in the home directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied JSON files (e.g., endpoints.json, study.json) to calculate scores and estimates. While this presents an attack surface where malicious data could influence agent-derived conclusions, the impact is minimal as the processing is performed by deterministic scripts rather than being directly interpolated into LLM prompts.
Ingestion points: User-provided JSON data files read via the --input and --target flags in the assessment scripts (endpoint_selector.py, phase_gate_scorer.py, ar_evaluator.py).
Boundary markers: None; the scripts assume the JSON input follows a specific schema but do not explicitly verify the intent of the data.
Capability inventory: File system read/write (limited to local configuration and input files), diagnostic reporting to stdout.
Sanitization: Standard json.load() parsing with type casting for numeric values; no use of eval or exec on input data.

clinical-research