jev-model-routing
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted inputs such as user tasks and prompts to make routing decisions. * Ingestion points: Processes user-provided text through --prompt and --task arguments. * Boundary markers: No specific boundary markers are defined to prevent the agent from following instructions embedded in the task data. * Capability inventory: Possesses shell execution capabilities via the jev CLI tools. * Sanitization: While it includes a secret redaction mode, there is no broader sanitization for malicious prompt instructions.
- [COMMAND_EXECUTION]: The tool supports running arbitrary shell commands via the --run flag in jev lane step, which can be dangerous if input commands are influenced by external data.
- [DATA_EXFILTRATION]: By design, the skill sends turn content to an external classification service ('Jev'). Although it attempts to redact secrets, this constitutes data exposure to a third-party endpoint.
Audit Metadata