Guides
Brief like a conductor.
Six Claude Code sessions multiply good briefs and merely parallelize bad ones. Here is how to write goals that split well, size tasks that verify cleanly, and know the rare moment to type into a terminal yourself.
Writing briefs
State outcomes with seams.
The Conductor plans your goal into a mission DAG and dispatches tasks across the fleet. A brief splits well when it names independent pieces of work, and splits badly when every step depends on the last. Describe what done looks like and where the natural seams are, then let the planner cut along them. The Mission DAG page shows what a well-split goal becomes.
# splits poorly: one serial chain
Refactor the auth module, then update every
caller, then fix whatever tests break
# splits well: independent, checkable outcomes
Add request validation to all six API route
groups. Each route group gets its own tests,
and the full suite passesSizing tasks
One verifiable outcome per terminal.
Too small wastes the fleet
Six sessions renaming six variables is overhead, not parallelism. If a task would take you minutes in one terminal, fold it into a larger piece of the same brief.
Too big defeats verification
Loom re-checks work before it counts as done. A task sized to one module, one route group, or one package gives that check something crisp to confirm. "Improve the codebase" gives it nothing.
Right-sized tasks compound
One package per session is why monorepos, test sweeps, and upgrade missions work so well. The seams already exist. Your brief just has to point at them.
When to jump into a terminal
Mostly, you should not have to. Loom answers permission prompts, nudges drifting sessions, relaunches exited CLIs, and recovers rate limits on its own. The live activity strips tell you where every session stands without opening a single pane.
But every pane is a real Claude Code TUI, and you can type into any of them. It is worth doing when:
- You have context the Conductor lacks, like a flaky test you recognize on sight or a design call only you can make.
- You want to tighten or loosen one session's autonomy. Shift+tab cycles that terminal's permission mode, and auto-accept can be toggled, so one careful pane can run stricter than the other five.
- A session keeps circling a problem. Smart steering will nudge it, but one sentence from you often beats three automated nudges.
Jump in, say the thing, and step back out. The mission keeps running around you.
Hand it the work.
Walk away.
macOS, Linux, and Windows. Around 13 MB. Free and open source.