trace-to-training-data
Pass
Audited by Gen Agent Trust Hub on Jul 14, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted production logs and evaluation traces to generate training data, which creates a surface for indirect prompt injection. If the source logs contain malicious instructions, they could be inadvertently incorporated into the training datasets.
- Ingestion points: Untrusted data enters the context from
runs/<run-id>/results.jsonand human-provided corrections. - Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the processed log entries.
- Capability inventory: The skill performs JSON data transformations and filtering in Python. It does not invoke subprocesses or execute arbitrary commands based on the data content.
- Sanitization: The skill explicitly requires a 'secret and PII scan' and redaction process (Hygiene section), but it does not specify filters for adversarial prompt injection content in the raw data.
Audit Metadata