Scripted agent teams in Claude Code with Jev as the decision layer
Opus writes the workflow script, Sonnet runs in parallel, Fable advises, and mechanical decisions go to Jev in 16ms.
Claude Code tip: stop burning Opus 5.5 context coordinating 40 teammates turn by turn on tasks Sonnet 5.5 can swarm, while Fable 5.1 sits idle spin up a scripted agent team with /effort ultracode run /effort ultracode Opus 5.5 acts as the lead architect, writing the team's workflow script and reading only the final answer Sonnet 5.5 teammates fan out up to 16 at a time, each in its own isolated copy of the repo, at medium effort Fable 5.1 sits on call as the advisor every teammate inherits, only stepping in at three critical moments: → before the script runs: is this the right way to split the team? → when a teammate fails twice: is the script wrong, or the file? → before the report lands: which findings didn't survive cross-checking? Sonnet 5.5 builds. Fable 5.1 challenges. Opus 5.5 merges and ships Jev engineering runs the identical pattern one layer down: mechanical decisions that need no reasoning (which file to open, which tool to invoke, retry or abort) execute in 16ms, so the frontier models only wake up when execution paths actually diverge Plan on high. Delegate on medium. Keep Fable on call. - the full team setup > Opus 5.5 on high writes workflow.js and only reads the final answer > agent() maps the codebase before anyone edits > pipeline() hands one file to one Sonnet 5.5 teammate > parallel() fires the adversarial verifiers at once > up to 16 teammates live, 1,000 per run, results stay in script variables > press s in /workflows and the whole team becomes your own /command paste the setup and this prompt into Claude Code below: "Configure my Claude Code workspace for scripted agent teams: 1. Check ~/.claude/settings.json, .claude/settings.json and my env for anything that turns workflows or the advisor off > disableWorkflows, enableWorkflows set to false, CLAUDE_CODE_DISABLE_WORKFLOWS > CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY > Report them, change nothing 2. Lock the lead in ~/.claude/settings.json: > Set effortLevel to high and advisorModel to fable > Set workflowSizeGuideline to large 3. Add a team rule to ~/.claude/CLAUDE.md: in workflows, run read and edit stages on sonnet at medium effort, give each editing agent its own isolated copy, and put every finding through an adversarial verify stage before it reaches the report 4. List the folders in this repo where a one-agent-per-file team would pay off, with an estimated teammate count for each Show every configuration diff first. Do not apply edits until confirmed" save it
This post quotes another post: x.com/eng_khairallah1/status/2107767577963594051