Claude Code 多 agent 树：Opus 规划、Haiku 并行、Jev 处理确定性分支

Grirk @Grirku原推发布于 2026.10.22

在 X 上打开原推

THIS IS A PURE F*CKING GOLDMINE Anthropic just showed one of the cleanest Claude Code tricks. Opus 5.5 writes the plan as a script, then sits idle while 3 Haiku 5.5 agents actually do the work. They run in parallel: → security audits jwt.ts and patches the timing leak → perf-scout profiles redis locks (p99 14.2ms → 3.8ms) → qa-engine runs the full auth test matrix (58/58 green) They coordinate through a peer p2p bus. No leader polling. Atomic mutexes on files. Self-claim tasks. Direct mailbox messages. JEV sits one layer down and resolves thousands of micro-forks (which file, which model, retry or stop) in code. Only the real splits ever reach Opus. Opus only appears twice: → writing the plan → signing off on the verified fixes Opus plans. Haiku scans. Sonnet ships. The full tree > Opus 5.5 runs the main session on xhigh and writes the plan > security audits on Haiku 5.5 > perf-scout profiles on Haiku 5.5 > qa-engine tests on Haiku 5.5 > peer p2p bus + atomic mutex DAG > Opus only reads what survived, then signs off > JEV handles deterministic forks in milliseconds Paste the tree + this prompt into Claude Code ↓ Rebuild my Claude Code setup around this tree: Run the main session on Opus 5.5 with effort xhigh and --teammate-mode auto. Have Opus write the plan as a script: security on Haiku 5.5, perf-scout on Haiku 5.5, qa-engine on Haiku 5.5. Let the phases overlap, each finding moves forward as soon as its own stage finishes, no waiting for the batch. Never surface raw agent output to the main session, only verified results. Route deterministic forks (model choice, lane choice, retries) through JEV. Show me every change as a diff first. No edits until I say go. save it, it’ll come in handy.

这条是引用着另一条写的：x.com/mirku21/status/2098836008468992471