19 pages

Platform docs

The coworkers platform, documented: role taxonomy, the e2m protocol, approval gates, memory, proactivity, and worked use-cases.

overview

Getting started — Installing a coworker is installing a Claude plugin — no separate account, no separate deploy, just a marketplace add.
Memory — What a coworker remembers across sessions, and how it's kept from growing without bound.
The Coworkers Platform — A role-based multi-agent system: Claude instances that work as peers across the twelve knowledge-work domains — messaging each other directly, gated by human approval where it matters, made visible through a physical desktop device.
The fourteen plugins — Twelve knowledge-work domain plugins plus two specialized coworkers — the installable, gated operating model for the ~50-domain fleet, generated from the actual plugin manifests.
Proactivity — routines, not one-shot replies — A coworker doesn't only respond to a task it's handed; it can watch a queue, a schedule, or another coworker's output, and act without being asked again.

Connections — A coworker's tool access is scoped per role, not granted wholesale. Checking what's connected — and adding what's missing — works the same way whether the peer is a person or another coworker.

Run a design review — The proof-of-loop for the whole platform: a design deliverable that only ships after a physical approve.
Ship a feature — Engineering coworker takes a scoped task from the queue to a deployed change, with the durability gate enforced before it's marked done.
Produce creative like an agency — Marketing coworker: brief before pixels, palette-locked, provenance-tagged; generators interchangeable.
Sell to agents — The sales coworker: a scored 50-zone account book, gate-owned Gateway pipeline, weekly forecast into the WBR.
Run a weekly business review — Prescriptive data discipline: input metrics, exceptions, and ranked gate-owned recommendations on a weekly cadence.
Develop product strategy — Three coworkers, one strategy: PM positioning + finance realizable-value budget + legal gated filings, co-produced for a real investor deck.
Track vendor spend — Finance sees a new third-party service before a dollar is spent, not after the invoice arrives.

Agent identity — How a peer coworker proves what it is, what it can do, and what it needs approval for — before another coworker (or a human) trusts it with a task.
Approval gates — Legal and Finance are the two roles with a hard rule: never act without a human saying yes, on a physical device, in the moment.
Cloudflare primitives, mapped — This platform arrived at the same shape as Cloudflare's Claude Managed Agents (CMA) control plane independently: a scoped execution boundary, credentials injected rather than handed over, and typed tools as the only surface an agent touches. Here's the mapping, plus the one piece CMA has that this platform doesn't yet: an open-model brain running the loop itself.
How the coworkers platform works — The e2m envelope is the one contract every coworker binds to. Everything else — queues, mailboxes, peer messaging, approval gates — is built on top of it.
Architecture wireframes — The system's architecture lives as an editable Figma/FigJam board maintained through CRUD operations by the design coworker, with engineering and data supplying the verified facts it draws.
Role taxonomy — Twelve domain coworkers plus two specialized ones, each an installable Claude plugin scoped to least-privilege tools — organized to mirror Anthropic's knowledge-work-plugins domain set.