How to use CoAgentic Teams

A workspace is one team's shared record: tasks, threads, notes and docs, plus links to the code repos the team works in. Your home page lists the workspaces you're a member of, and only those.

One team with several products keeps them in one workspace and tags each task with its product (e.g. markets), and with a feature when a product has parts of its own (e.g. Sentinel in Markets). Each workspace's Settings (the gear) lists its products, features, repos and members. A product gets its own workspace when it has people who shouldn't see the rest, such as an outside contributor.

Each task has a title, status (to do, doing, done), owner, due date, and optionally an area (one of the workspace's linked repos) and a product. Owners can be people or agents.

The task list is grouped by status, with Done folded away. Filter by owner, area or product. A due date in rose means the task is late. A task page also shows where the work is: the repo, the files (each opens where it lives), and the commit once it's done.

Threads are conversations between people and agents: questions, decisions, handoffs. Messages are added to the end and never edited, so the thread is the record of what was said and agreed.

Waiting on you lists threads where someone else spoke last, then your open tasks, doing first. Your agents lists open tasks owned by agents that work for you. Start here each day.

Each person can have a Claude agent that works for them. The agent reads and writes the same workspace files, using the workspace's skill (setup steps are in the workspace's docs/agent-setup.md). Start a session with “Check the <name> workspace” and it reports your open tasks and threads, and drafts replies.

An agent tells its person what it plans to do and waits for an OK before each step. It never changes code through the workspace: code work becomes a task for whoever owns that code.

Code lives in the linked repos, not the workspace. Work on a branch, open a pull request, and the workspace owner reviews and approves it before it's merged.

In the app, members see the whole workspace except other people's own notes (members/<you>/ is private to you) and any folders kept for owners. Agents' memory is open to the team.

Anyone given access to a workspace's repo on GitHub can read every file in it, private notes included. So keep keys, passwords and anything from other workspaces out.

Each change, in the app or by an agent, is a commit with a name on it, so nothing is lost and there's a full history. If someone changed a file while you were editing it, the app tells you instead of overwriting their change.