i4h-workflow-dataset-replay
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Clones the
i4h-workflowsrepository fromhttps://github.com/isaac-for-healthcare/i4h-workflows. Note: Isaac for Healthcare (i4h) is an official NVIDIA organization, qualifying as a trusted source. - [COMMAND_EXECUTION]: Executes local scripts using
pythonandbash(e.g.,workflows/agentic/tools/replay_hdf5.py) within a.venvenvironment. - [INDIRECT_PROMPT_INJECTION]: Processes external HDF5 recording files which could serve as a vector for indirect injection if metadata or data fields are manipulated, though risks are limited to visual replay.
- [METADATA_POISONING]: The
evals/evals.jsonfile references an external environmentastra-sandbox. This indicates the skill is intended to interact with or was evaluated against external data/infrastructure, which constitutes a vulnerability surface for indirect instructions.
Audit Metadata