Claude Code team quality gates: Jev runs mechanical decisions in 16ms

Khairallah AL-Awady @eng_khairallah1Originally posted 2026.10.06

Open original on X

Claude Code tip: stop letting agent teammates mark tasks done that Opus 5.5 has to redo, while Fable 5.1 sits idle load quality gates for the whole team with --settings claude --model opus --settings .claude/team-gates.json Opus 5.5 acts as the lead, writing the shared task list and only reading work that passed the gates Sonnet 5.5 teammates claim tasks from the list and can't close one until the gate lets them Fable 5.1 sits on the team as a dedicated adversary, and three hooks decide when work gets pushed back: → TaskCreated: does this task name the test that proves it? exit 2 sends it back → TaskCompleted: does that test pass right now? exit 2 keeps the task open → TeammateIdle: is unblocked work still waiting? exit 2 puts the teammate back to work 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 Lead on Opus. Gate every task. Keep the adversary on call. - the full team setup > Opus 5.5 on high leads and writes the shared task list > ux, backend and tests teammates on Sonnet 5.5 claim tasks, file locking stops double claims > TaskCreated rejects any task without a test command > TaskCompleted runs that command and blocks "done" on a failure > TeammateIdle hands idle teammates the next unblocked task > adversary on Fable 5.1 reviews each task before the lead synthesizes paste the setup and this prompt into Claude Code below: "Configure my Claude Code workspace for a gated agent team: 1. Create .claude/team-gates.json: > env CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 > a TaskCreated hook that exits 2 when a task description has no test command > a TaskCompleted hook that runs that test command and exits 2 with the failure output > a TeammateIdle hook that exits 2 while unblocked tasks remain 2. Write each hook script in .claude/hooks/, then dry-run every one against a fake event payload and show me the output 3. Create .claude/agents/adversary.md with model: fable and tools: Read, Grep, Glob, Bash 4. Add a team rule to ~/.claude/CLAUDE.md: every task description names its test command, and the adversary reviews each task before the lead synthesizes Show every configuration diff first. Do not apply edits until confirmed" (Link in the comments)

This post quotes another post: x.com/eng_khairallah1/status/2106681096893317172