Claude Code Multiplayer
Build One Shared Workspace for Your Team and Agents in 30 Days
Two agents just rewrote the same file and both reported done. Build the shared workspace that stops it: an owner on every file, a proof gate neither agent controls, a run log that names the first broken step, and your own measured number for one agent against the team.
Your agent finishes. Your teammate’s agent finishes. One of the two writes is gone, and nothing in the system has an opinion about it. No merge conflict, no error, no red text. It looks like success reported twice.
This book builds one shared workspace for a team and its agents, end to end: an owner on every file, a handoff format both agents obey, a gate outside both agents that refuses a false “done”, scoped access with an injection test you re-run after every change, and a run log that names the first broken step instead of the last one.
Every file ships in a free companion repo: thirteen shell gates, one folder per chapter, and fixtures so you can watch each gate refuse and then watch it stay silent.
What You'll Build
Reproduce the silent collision in five minutes: two agents, one file, one write gone, no conflict marker.
Score three of your own tasks on a split-test sheet with your measured cost floor, and reject at least one.
Build a workspace with an index, an owners file, and an inbox a second person can clone and use without a conversation.
Run a two-agent handoff on a six-field message format, with a validator that refuses a malformed one.
Wire a proof gate outside both agents that throws back a "done" on work that does not build.
Add a claim check on every edit that grants the file or refuses the second agent by name.
Give every agent a bounded scope, then attack your own workspace with a hostile handoff and log the refusal.
Log every tool call per agent, then name the step that started a broken run instead of the one that crashed.
Price your last run step by step, name the most expensive one, and build a rate card from it.
Run the revocation test that proves the agent forgets a teammate who left, then hand the next hire a rollout pack.
Schedule a sweep that flags the stale notes every agent still reads, with the next sweep date already written in.
Run the paired benchmark on your own work: one agent against the shared workspace, verdict written down.