Pipeline Kit
Enterprise teams already have coding agents. They do not have a repeatable factory: the right workflow, the right model, and a pack that survives the next customer. Pipeline Kit is that operating model.
Same process on every engagement. Overlay the customer in config, not in skills.
Classify the step, then pick the model. Reasoning models plan. Coding models implement.
Knowledge, plugins, and traces stay opt-in. The core pack stays portable.
Enterprise factories
The work is making agents behave like a plant: one procedure, measured cost, and a model that matches the station. Scroll the left column. The answer stays pinned.
Factory controls
Not a catalog of features. These are the controls operators actually turn: which model, which knowledge, which plugin, and whether the run is scored.
Model classifier
Jev does not invent the next agent. The sealed graph does that. Jev only answers: which Cursor model should run this step.
This step
Planning — requirements and acceptance criteria
Runs claude-opus-5
Reasoning. Composer is a poor fit.
Two runtimes
Kit mode — default
Skills, briefs, and the loader live in .pipeline. The IDE only exposes run-workflow. This is what most engagements should use.
Orchestrator mode
Sealed graph. Drive it with run, approve, and resume. Pass --request. Jev classifies the model before each agent.
uv tool install git+<repo-url>
cd /path/to/your-app
pipeline-kit init --ide cursor
pipeline-kit doctor --ide cursorSubscription
Install stays open. One license turns on the controls a delivery org runs across engagements.
Included
Kit mode, ask, and feature-development. Loader, wiki, planning gates, and policy hooks. Install, doctor, and the workflow list stay open.
Org license
One license covers every engagement on the machine. Activate with pipeline-kit license activate.