Four Claude Agents That Work Like a Dev Team

Turn Claude into a four-agent dev team that ships real features while you sleep. Set it up once, then run the whole thing from one command

Here is the whole thing I could not fit in the video. You set up four agents, each locked to ONE job, that hand off through a single shared folder. One plans, one builds, one tries to break it, and a read-only reviewer guards the gate before anything touches your real project.

The one-time setup

Do this once. It takes about 15 minutes.

1 Make one shared folder (or repo) the whole team works in. Everything they produce lives here.

2 Create four agents, one per role below (separate Projects, subagents, or saved instruction sets).

3 Paste that agent's prompt (below) into its instructions. This is what locks it to one job.

4 Wire the handoff: each agent writes its output to a file in the shared folder, and the next one reads it. Planner writes spec.md, Coder writes the code, Tester writes test-report.md, Reviewer writes the verdict.

5 To run the whole pipeline from one command, save them as subagents or a slash command that fires them in order: plan, build, test, review.

Your 4 agents

Copy each prompt straight into that agent. The order is the whole point: each one only trusts the file the last one left.