Skip to content

Multi-Claude Handbook

Multi-Claude is a lane-based parallel build system for Claude Code. It turns large tasks into packet graphs — small, independently claimable units of work with explicit file ownership and dependency edges.

This handbook covers everything an operator needs to plan, execute, observe, and close parallel runs.

You’re an operator managing parallel Claude Code sessions across a codebase. You need to know:

  • When multi-claude is the right tool (and when it isn’t)
  • How to plan and freeze a blueprint
  • How to monitor execution and intervene when things go wrong
  • How to close a run and generate evidence for review

Every run follows the same six-step loop:

  1. Plan — Assess fitness, generate a blueprint, freeze the contract
  2. Execute — Workers claim packets and produce artifacts
  3. Observe — Watch the live console for state, hooks, and fitness
  4. Intervene — Stop, retry, resolve, or approve as needed
  5. Recover — Use guided flows for the 8 known failure scenarios
  6. Close — Derive outcome, generate handoff evidence, promote or reject