atai-task-verification-agent
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied video recordings and reference procedure text files to be processed by a remote AI agent. This establishes a surface for indirect prompt injection if the ingested content contains adversarial instructions.
- Ingestion points: The
references/run_tva_agent.pyscript reads local file paths provided as arguments and uploads their content to the agent platform. - Boundary markers: No programmatic delimiters or specific 'ignore instructions' warnings are added to the procedure text before it is submitted to the API.
- Capability inventory: The skill possesses the capability to upload data, execute verification runs, and download output files via the vendor's API.
- Sanitization: The procedure text is processed and uploaded without sanitization or escaping of potential embedded instructions.
- [EXTERNAL_DOWNLOADS]: The skill utilizes the
archetypeaiPython package and communicates with theplatform.u1.archetypeai.appAPI for processing. It also retrieves results from presigned URLs provided by the vendor API. These resources and endpoints belong to the official infrastructure of the skill author and represent legitimate tool operations.
Audit Metadata