roboflow-training-and-evaluation
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides instructions for the agent to ingest and act upon data retrieved from the Roboflow platform, which represents an indirect prompt injection surface if the external platform data is manipulated.
- Ingestion points: The agent retrieves untrusted data from the platform through tools like
model_evals_get_image_predictions(referenced inSKILL.mdandmodel-diagnosis.md),projects_health, andmodel_evals_get. - Boundary markers: The instructions lack explicit directives for the agent to use delimiters or boundary markers (e.g., xml tags) to isolate this external data from its primary instructions.
- Capability inventory: The skill enables high-impact capabilities such as creating new training runs (
trainings_create) and generating dataset versions (versions_generate) based on the evaluation of ingested data. - Sanitization: No specific data validation, escaping, or sanitization procedures are defined for the agent to apply to the content retrieved from the model evaluation tools.
Audit Metadata