Splitting work across Opus, Sonnet and Fable with Jev underneath
The author splits Claude Code into layers: Sonnet does the work, Fable acts as an outside advisor, and Jev handles small branching decisions, with a paste-ready setup prompt included.
This is f*cking insane. I spent thousands of dollars before I split this across three models. Run Opus 5.5, Sonnet 5.5, and Fable 5.1 together, so Opus stops doing work it was never built to do: Sonnet handles three roles, reading, editing and testing, researching, and Fable watches the session from outside, silent until one of three moments hits. where Fable actually steps in: → right before a plan goes live: is this genuinely the right move? → the exact same error shows up a second time: is the search going nowhere? → a long task gets closed out: did something get missed? a layer under that, Jev engineering catches the smaller forks: which file to touch, which tool to run, push forward or stop. none of that calls for real reasoning, so none of it belongs on a model priced for real reasoning. if one model is still answering every single decision in your session, you're spending Opus-level money to confirm a file is actually there. the exact setup, paste this straight into Claude Code: "Rebuild my Claude Code setup around this structure: Look through ~/.claude/agents and .claude/agents for subagents already covering explorer, worker, and researcher. Only create new ones for roles that are missing. Set model: sonnet, effort: medium on each. If an existing subagent is locked to a different model, leave it as is and just list it. In ~/.claude/settings.json, set effortLevel to high and advisorModel to fable. Check for anything disabling the advisor, CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, or anything blocking feature-flag fetches, plus CLAUDE_CODE_EFFORT_LEVEL, which can override subagent effort settings. Report what you find. Don't change any of it yet. Add one line to ~/.claude/CLAUDE.md: check in with the advisor before a big plan, when the same error shows up twice, and before marking a long task done. Show every change as a diff first. Wait for my go-ahead before touching anything." http://code.claude.com/docs/en/advisor
This post quotes another post: x.com/rvaniaaaa/status/2101709732314558639