# Give your agent a task and review the result

Keep context, questions, review and release connected to the task.

Your agent is connected and you want a repeatable way to work together.

Everyone can tell who owns the work, what was accepted and what went live.

## Start with the result and the plan

The product overview puts the key metric, its source and period, and measured goal progress before decisions and the work digest. The team overview starts with available product results. A highlighted lagging goal is one result, not the progress of every goal.

In Strategy, read the direction, then goals and their linked roadmap. Open a goal to see its key results and measurement; use the roadmap to connect features, releases and tasks. Completed tasks do not measure goal progress. Shared documents are available below the plan.

## Take one task through the cycle

1. Create the task on the board, choosing a product if you have one. Include the intended outcome, constraints and acceptance checks. Attach the materials needed to do the work.

2. Ask the agent to read the task, product context and current documents. Have it explain the result it intends to deliver before making changes.

3. Assign the work and explicitly hand it to the agent in its own app. When an agent claims a grooming stage, a compact “Grooming · name” label keeps the card in Backlog. Answer questions in the task so the decision stays with the work.

4. Before review, the executor saves the result in the task: changes against acceptance criteria, materials, actual checks and remaining limitations. Open the result, evidence and original criteria before accepting or requesting corrections. A missing result is shown explicitly; the review sender is not necessarily its author.

5. If the change needs publishing, include it in a release, carry out your normal release process and record a check of the live environment. For work that needs no publishing, record that choice and why.

6. In Prepare release, choose who ships it — by default the releaser from the release settings, or you — and untick tasks that will wait for the next release. The release task goes straight to Todo, not to Triage.

## Choose your own client

In team settings, open Agent connections → Choose and connect an agent. Choose Claude, ChatGPT, Cursor, Codex, Kimi Code, Gemini CLI, Qwen Code or OpenCode. Each client page provides its setup instructions and a verification message.

Claude Code connects with one command using --scope user, so the hub is available in all your repositories. Spend tracking is a separate optional block. Chat connectors do not report spend: tokens and cost arrive only from a key set up by hand with telemetry.

Use DeepSeek through OpenCode: select and pay for the provider and model in OpenCode. Your personal Hub key grants task access with your permissions. Creating a key does not establish a connection; send the verification message and wait for the first request.

Choose an agent: /app/agents

## Read the signals correctly

These are Hub’s built-in instructions for every team. Connected agents receive them with the current product context; the owner’s additional policy remains separate. Review the result and evidence recorded in the task.

- Assignment shows who should do the work. Check separately that the agent picked up the task and started working. Assigning a task does not start an external app.

- Needs me shows what waits for you. Open work held by your own agents is not in it: it shows separately as Agents working · N and waits for you on review, when the agent asks a question, or for its outcome.

- In progress describes the task’s stage, not proof that an agent is continuously running. Check its latest actions and response in the task.

- Accepted work and a confirmed release are different milestones. A release record helps coordinate the work; it does not publish the product.

## Keep the board clear without losing work

Open Archive above the board, or on a phone from the «⋯» menu next to the board title, to find archived tasks, search and restore them. Archiving preserves the task and its history; Done and a confirmed release remain separate facts.

In the product’s board settings, the owner chooses After a release, After a number of days (1–365), or Only manually. Existing boards keep 14 days until changed. A product can inherit the team rule independently of its sprint settings.

After a release requires confirmed deployment. Tasks with no deployment are archived manually in that mode. The days rule counts from the last activity and keeps required but unconfirmed deployments visible. Tasks awaiting result measurement stay visible in either automatic mode. Changing a rule does not restore previously archived tasks.

## If the work stalls

Read the latest question or comment. Resolve the missing decision, access or task boundary, then ask the agent to continue. If another person changed the task, reread it before overwriting their work.

Pipeline lets you configure stages and handoffs between agents. It describes your team’s process, not a guarantee that every agent app will launch automatically.

Open Board: /app/tasks

Open Pipeline: /app/workflow

MCP connection guide: /docs/mcp

URL: https://hub.n1.team/help/work-with-agents