validate-structured-model-output
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external model outputs in JSONL format, which constitutes a potential indirect prompt injection surface. 1. Ingestion points: Data is read from local files specified via command-line arguments in scripts/validate_jsonl_outputs.py. 2. Boundary markers: The skill documentation explicitly instructs the agent to treat model output as untrusted input and provides a multi-stage validation framework to isolate different types of failures. 3. Capability inventory: The skill performs local file system reads and writes to generate validation reports but lacks network or dynamic execution capabilities. 4. Sanitization: The tool performs strict JSON parsing, type validation, and identifier checks, which serves as a security control for subsequent data processing.
- [DATA_EXPOSURE]: The validation script accesses local files to perform its function. This behavior is restricted to the specific input and output paths provided during execution and does not include access to sensitive system directories or credentials.
Audit Metadata