Skip to main content
Everything here ships in the binary today. The lifecycle — spec → build → evaluate → diagnose → optimize — has its own page; these are the capabilities around it.

Sub-agents

Dispatch agents in parallel, then watch and steer them live.

Agent mode

A lean coding agent on the same harness, without the lifecycle.

Goals

A loop that runs until the goal lands, blocks, or spends its budget.

Monitor

A background watch on a long command, streamed as notifications.

Trace viewer

/trace — a read-only walk through your session logs.

Prime

An agent loop whose only tool is code it writes and runs. Experimental.

Project context

AGENTS.md / CLAUDE.md, loaded into every agent’s prompt.

Session commands

The slash commands that drive a running session.

How the loop works

The five lifecycle stages these capabilities serve — and the two rules that keep the loop honest.