# Collaboration: in a group, between groups, and with you

> How Agents pass work: handoffs, wake-ups, shared files, TaskBoards and sub-Graphs in a group; boards and messages between groups; the main assistant and you.

Agents in The One pass work and information in several ways, each suited to a different kind of exchange.

## Inside a working group [#inside-a-working-group]

1. **Handoff.** An Agent calls `handoff_work` to pass work, files or a request to an Agent it is connected to by an edge. The sender goes to sleep and the receiver wakes with what it was handed, and nothing else of the sender's context.
2. **Parallel wake-up.** With `activate_node` an Agent wakes another Agent of the same working group with what it should do, and carries on itself without waiting for it. This is how the shipped team's checkers start alongside the lead when the work begins.
3. **Shared files.** The Agents of a group work in the same folder and read what the others wrote.
4. **TaskBoard.** An Agent plans tasks in layers. Each task gets a worker and a reviewer (the reviewer is cloned from the Agent that started the board), the tasks of one layer run in parallel, and the Agent gets the results back when the layer is done.
5. **Sub-Graph.** An Agent runs another template as a sub-Graph and acts as its human: the sub-Graph's questions come to it, and it answers with `reply_to_subgraph`.

## Between working groups [#between-working-groups]

6. **The matter board.** Every group on a matter reads and writes its cards: the facts, decisions, checklists, links and files that others will need. Reading the board wakes nobody.
7. **Messages between groups.** With `message_working_group` a group writes to another group on the same matter or on a matter above, below or beside it: to ask, request, reply, or say what a reply still lacks (rework). A request reaches the other group once it is resting, and that group decides whether and how to take it up.
8. **Sub-matters.** A group can open part of its work as a sub-matter, with its own board and its own working group.

## With you [#with-you]

9. **The main assistant.** It briefs a group with your own words, passes on your follow-ups, and answers a group's questions or leaves the ones that are yours to decide to you.
10. **You.** When an Agent needs you, it asks: the question shows in the strip above the conversation (**2 waiting for you**, say) and on the matter's board. You can also talk to any working group directly.

## Reading and editing another Agent's context [#reading-and-editing-another-agents-context]

Every Agent manages its own context: it expands and collapses its own Skills and files, or meditates. Some Agents also hold the context-editor tools (`context_editor`): they can read and change any Agent's context by its id, to correct what it has understood and help it work better. In the official Space the main assistant, the general `chat` group's assistant and the nightly review's Reviewer hold them. The host first finds the friction that keeps recurring in the records; the Reviewer then uses these tools to read the other Agents' histories around the time it happened, to work out where it comes from. Any role can be given them in its editor.

Source: https://the-zeroth.com/en/docs/multi-agent-architecture/collaboration