jev-triage
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted text (e.g., support tickets or inbox messages) for classification, which presents a surface for indirect prompt injection attacks.\n
- Ingestion points: Record text is ingested into the
state.textfield of the classification request (as seen inassets/example.json).\n - Boundary markers: The skill relies on structured JSON fields but lacks explicit delimiters or instructions for the model to ignore commands embedded within the message content.\n
- Capability inventory: The skill uses a CLI tool to interface with an external API for triage but does not perform automated mailbox modifications.\n
- Sanitization: No explicit sanitization of the input text is documented.\n- [EXTERNAL_DOWNLOADS]: The skill requires the
jev-skillPython package and thejev-decideCLI tool.\n - Evidence: The documentation refers to the requirement for the
jev-skillpackage andjev-decidetool to be present in the environment.\n - The instructions correctly advise the agent to warn the user about missing dependencies rather than performing an automated, silent installation.
Audit Metadata