Agent View

Agent View

Use agent view to dispatch, monitor, resume, and manage many Claude Code background sessions from one screen.

What Agent View Is

Agent view is a control panel for multiple Claude Code sessions. It shows what is running, what needs input, what finished, and what failed. Each background session is a full Claude Code conversation that can keep working without a terminal attached.

Open it with:

bash
claude agents

Do not confuse this with the /agents in-session command for custom subagents. claude agents manages sessions. /agents manages subagent definitions and running subagents inside a session.

What You Can Do

Agent view helps you:

  • Dispatch new background sessions.
  • Watch session state without reading full transcripts.
  • Peek into a session and reply when it needs input.
  • Attach to a session when you want to take over.
  • Filter sessions by repository, branch, status, or pull request.
  • See pull request status for sessions that create PRs.
  • Clean up old background worktrees.

It is the right tool when you are supervising many independent Claude Code sessions rather than one team lead with teammates.

Agent View vs Agent Teams

QuestionAgent viewAgent teams
What is managed?Independent background sessionsTeammates coordinated by one lead
Do workers talk to each other?No, you coordinate themYes, via task list and mailbox
Is there a shared plan?Only what you imposeYes, team lead manages it
Best forMany independent tasksOne complex task with peer coordination

Use agent view for portfolio management. Use agent teams for collaboration inside one task.

Dispatch Patterns

From agent view, dispatch focused prompts with clear ownership:

text
Review the payments API for missing validation.
Do not edit files. Return file paths, severity, and suggested fixes.

For implementation sessions, give each one a narrow lane:

text
Implement the settings page loading state only.
Do not modify API routes or auth code.
Open a PR when tests pass.

File Isolation

Background sessions need file isolation when they edit code. Claude Code can use worktrees so each session has its own branch and working directory.

Good background tasks are independent:

  • Review one subsystem.
  • Fix one failing test group.
  • Implement one small feature.
  • Draft one migration plan.

Avoid dispatching several background sessions that edit the same files unless they are research-only.

When Not to Use Agent View

Stay in one foreground session when:

  • You need rapid back-and-forth.
  • The task is small.
  • Several subtasks depend tightly on one decision.
  • You need to watch every tool call before it happens.

Use a dynamic workflow when you want one run to spawn many agents automatically. Use agent view when you want to manage separate sessions yourself.

Official References

Continue with practice

You have finished the core ideas of Agent View.

If you want to turn the idea into something reusable, continue practicing on AgentWay.