> ## Documentation Index
> Fetch the complete documentation index at: https://docs.usesyndicate.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Working with Atlas

> How the Manager plans, delegates, reports, and asks for your input, and how to tune all of it.

Atlas is the Manager agent at the center of every team. You give Atlas goals; it plans phases, creates tasks, dispatches specialists, keeps the board clean, and escalates decisions to you. Atlas coordinates rather than implements: it inspects the project read-only and delegates the actual work.

## Talking to Atlas

Open the team's **Chat** tab. Give Atlas the goal plus constraints, and say what done looks like:

> "Ship a pricing page this week. Use the existing design system. Frontend builds it, QA verifies it on mobile, and check with me before changing the nav."

### Quick Briefs

Ready-made prompts available in Atlas chat for common asks:

| Brief                | What you get                                              |
| -------------------- | --------------------------------------------------------- |
| **Status brief**     | Current work, recent changes, blockers, and next move     |
| **Phase / plan**     | Current phase, goal, owners, checkpoint, decisions needed |
| **Needs from me**    | Decisions, approvals, access, or missing inputs           |
| **Blockers**         | Only things stopping or slowing the team                  |
| **Next best action** | One recommended user action, with why                     |
| **Ship check**       | Release readiness, blockers, verification, confidence     |

### Slash commands

Atlas carries the Manager skill set. The most useful in chat:

* `/think`: pressure-test an idea before building starts
* `/delegate`: break a goal into tasks and assign specialists
* `/status`: pull progress, blockers, completion state
* `/pipeline`: run strategy, design, and engineering reviews
* `/launch`: run the shipping workflow (tests, review, changelog, PR)
* `/safeguard`: tighten safety controls before risky operations
* `/scope`: keep agent work constrained to intended files
* `/reflect`: run a retrospective
* `/learn`: save important context and decisions to team memory

## Assigning work yourself

You do not have to route everything through conversation. Open the slash menu and choose **Assign task**: pick the agent, describe the work, and choose how it starts: **Notify** (sends the work without starting immediately), **Start** (begins immediately), or **Flag** (waits for your decision).

## Where your attention goes

Atlas routes everything into three panels (full detail in [Tasks](/features/tasks)):

* **User Tasks**: decisions, approvals, and questions waiting on you. You act on these.
* **Agent Tasks**: the work board. You watch; you rarely need to act.
* **Task Health**: Atlas's cleanup queue. Glance at it if clutter piles up, then **Nudge Manager** or **Sweep stale tasks**.

In-chat **interruption cards** appear when Atlas wants to message an agent or open new work and your approval is required: **Allow & Send** / **Reject** for messages, **Allow New Work** / **Reject** for new workstreams.

## Interruption Level

Team Settings includes an **Interruption Level** that controls how often Atlas stops and asks you:

| Mode         | Behavior                                                                                                                   |
| ------------ | -------------------------------------------------------------------------------------------------------------------------- |
| **Minimal**  | "Atlas only interrupts for blockers, explicit @user requests, high-risk work, missing access, or required approval."       |
| **Balanced** | "Atlas coordinates routine work freely and interrupts for unclear scope, meaningful new workstreams, or medium/high risk." |
| **Hands-on** | "Atlas interrupts before major new assignments, broad follow-ups, risky decisions, and completion reviews."                |
| **Manual**   | "Atlas turns most delegations, follow-ups, and re-routes into approval prompts before agents continue."                    |

Start with **Balanced**. Drop to **Minimal** once you trust the team on a project; raise to **Hands-on** or **Manual** for risky or unfamiliar work.

## Updates from the team

The **Show updates from agents** checkbox in Atlas chat controls whether short "Update from \[agent]" digests appear inline as teammates report progress. The setting is remembered per team.

## Use Atlas well

* Give the goal and the constraints, not a step list. Atlas plans better than it follows micro-instructions.
* Use Quick Briefs instead of scrolling history when you return to a team.
* When the chat gets noisy, ask Atlas to restate the current plan in one message.
* If Atlas seems stuck or stale, the **Guide to Atlas** button (question mark in the chat header) summarizes all of these mechanics in-app.