jev-control-plane
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONPERSISTENCEDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes local Node.js and Bash scripts (e.g.,
jev-harness.mjs,jev-setup.sh) located in the~/.agents/jev/directory to perform configuration and status checks. - [DATA_EXFILTRATION]: It reads local configuration files such as
agents.jsonand transmits task context or code proposals to external APIs (e.g.,api,ollama) for policy decisions. - [INDIRECT_PROMPT_INJECTION]: A risk surface exists when the skill processes untrusted task data and code blocks via
prune-contextandreviewoperations. Maliciously crafted inputs could attempt to influence the decision engine's authorization results. - [PERSISTENCE]: The skill can install persistent hooks (e.g.,
hooks-install) and generate extension files likeextensions/jev-control-plane.jsthat are automatically loaded by the agent host. - [DYNAMIC_EXECUTION]: It supports loading and evaluating external skill instructions and JavaScript/TypeScript extensions at runtime, particularly when the
assistruntime mode is enabled.
Audit Metadata