Four different integration jobs
| Component | What it changes | What it does not establish |
|---|---|---|
| Agent skill | Gives the coding agent documentation for writing TypeSafe integrations. | It does not automatically insert Jev into every tool call. |
| Decision tool | Lets the agent ask bounded questions about supplied context. | It does not generate code or grant execution permissions. |
| Model or skill router | Chooses among available models, handlers or skills. | The selected LLM still performs the generative work. |
| Permission or context hook | Judges proposed actions or selects context to retain. | Model output alone is not an access-control boundary. |
The official coding-agent introduction makes the distinction explicit. Start with the TypeSafe agent skill when your goal is to write an application using the API. Inspect a community router or hook when your goal is to change the coding workflow itself.
Jev with Codex
For Codex-related integrations, first decide whether you need implementation guidance, a callable judgment tool or an approval hook. jev-code documents using Jev as a tool inside several agents, including Codex. OpenRouter's permission cookbook documents a different job: checking pending shell requests while retaining the host's permission flow.
Keep authentication, hook event names, host version and failure behavior tied to the specific integration. A plugin that supports one host or release should not be assumed to support another because both use the same model.
Jev with OpenCode
A generative OpenCode model and a Jev decision plugin occupy different parts of the workflow. The official TypeSafe guidance explains why selecting jev-latest as a generic coding model does not turn it into a code generator. A plugin can instead use typed decisions to route work or assist with permissions.
The OpenRouter permission recipe supplies a concrete example with static checks before model evaluation. Follow its current host prerequisites and configuration rather than copying a generic plugin registration from a differently named repository.
Hermes skill and tool selection
Hermes Jev Skills documents several components: model routing, skill selection, memory/context decisions and computer-use helpers. Pick the component that addresses your current bottleneck instead of enabling every capability together.
The official skill-suggestion recipe demonstrates shortlisting and checking candidate skills. Your code still owns the available catalog, the no-suggestion outcome and execution permissions. For a project-by-project breakdown of skill selection, approvals and routing, see Jev for Hermes.
Pi and Oh My Pi
omp-jev-router combines a Jev capability classification with provider usage limits to choose a model or delegated agent. That is a routing decision. y0usaf/pi-jev and pi-jev-auto-mode document tool-call checks, with their own configuration and failure behavior.
“Auto mode,” “watchdog” and “router” are not interchangeable. Read whether the extension adds a gate, preserves an existing gate, silently disables itself without a key, or refuses actions on an outage. Those differences determine the operational behavior. See Pi and Jev for the tool gates, auto mode and routers as separate projects.
Other hosts, frameworks and project names
Paperclip, LangGraph, Copilot-related integrations and Jev Kit projects can apply the same decision pattern at different boundaries. Check each project's evidence for the supported host and purpose. A framework router selects a path in an application graph; it does not automatically configure a coding CLI.
Evaluate one component in observation mode where available. Compare its proposed decisions with what your workflow actually needed, then introduce bounded actions with an explicit review path. Track retries, overhead and completion quality, not just how often an expensive model was avoided.