Starting a run
The run launcher is the Workbench's default surface. Describe a task, pick a provider, and dispatch — Nexus handles the worktree and session.
The run launcher
The launcher sits front and center in the Workbench. You write the task in plain language, choose how it runs, and hit Start run. Nexus creates an isolated worktree, opens a live session, and streams output.
Picking a provider
A row of provider chips lets you choose per run — Claude Code, Copilot, and (soon) others. The same project can run different agents on different tasks. Dimmed chips are providers that aren’t available yet.
Quick starts
For common intents, one-click chips pre-frame the task:
- Fix a bug
- Add a feature
- Write tests
- Investigate
- Refactor
- Document
Task templates
Reusable templates give you a structured starting point, tagged by kind:
| Template kind | Tag |
|---|---|
| Feature work | feature |
| Test coverage | test |
| Refactoring | refactor |
| Bug fix | bug |
Set up this repo
Before the first run, the launcher shows a readiness checklist for the selected repo:
- Branch detected — the current branch is recognized.
- CLAUDE.md — present at the repo root to orient agents.
- Worktrees configured — isolated working directories are ready for parallel runs.
A populated CLAUDE.md measurably improves agent output. The launcher flags its absence so you don’t dispatch a run blind.