What's this about?
Chinese KOL Baoyu shared his working pipeline for complex tasks: have a top-tier model (Fable 5) write the technical plan as a document, /compact the session, hand execution to a cheaper model (Codex), then bring the planner back to review the result.
What exactly is /compact?
A Claude Code command that compresses conversation history into a summary, freeing context space. Baoyu compacts immediately after the plan is agreed — the expensive context (the discussion) is distilled into the document, leaving room for verification later in the same session.
Why does it matter?
367 likes, active replies asking how-to. Multi-model division of labor — expensive model for judgment, cheap model for labor — is becoming the standard cost-control pattern, and the document (not the chat) is the artifact that travels between agents.