黑洞智能体
黑洞智能体BLACK HOLE AGENTS
Docs

Work / Agents

Work (also called Agents) is the system in BLACK HOLE AGENT that can call tools on its own, read and write files, and complete tasks across multiple steps — it's not the same thing as a chat assistant, which is just a role preset. You'll find it under Work in the left navigation bar (not the top tab used in older tutorials).

Choose an Agent, model and working directory in Work before starting; no working directory is selected in this example. (Chinese interface.)Choose an Agent, model and working directory in Work before starting; no working directory is selected in this example. (Chinese interface.)

Choose an Agent, model and working directory in Work before starting; no working directory is selected in this example. (Chinese interface.)

What It Can Do

  • Read and write files: give it a working directory and it can read, edit and create files there.
  • Call tools: built-in tools, plus the Skills and external MCP tools you attach.
  • Multi-step reasoning / subtasks: break down a goal, dispatch sub-agents, and run background commands.
  • Plug into automation: deploy it to IM platforms with Channels, or run it on a schedule with Scheduled Tasks.

Quick Start

  1. Open Work and choose an Agent. If you don't have one, click Create Project and create one in four steps (Basic info / System Prompt / Skills / Knowledge). The built-in Black Hole Assistant is ready to use.
  2. If the task involves local files, choose a working directory; otherwise the default workspace is fine.
  3. Describe the task in terms of "what should be delivered, which materials can be used, and what counts as done".
  4. Follow status / files / subtasks / message stream in the right panel; with developer mode on, you can also see the trace.
The Create Project wizard: name the project and choose a runtime, permissions and model. (Chinese interface.)The Create Project wizard: name the project and choose a runtime, permissions and model. (Chinese interface.)

The Create Project wizard: name the project and choose a runtime, permissions and model. (Chinese interface.)

Permission Modes

When an Agent works with files or commands, you can choose one of five permission modes, from most cautious to most hands-off: Ask Before Acting / Plan Only / Auto-accept Edits / Approve for Me / Full Access. Use the more hands-off modes only for unattended runs (such as Channels or Scheduled Tasks).

The permission-mode menu in Basic settings; choose the confirmation behavior for the task. (Chinese interface.)The permission-mode menu in Basic settings; choose the confirmation behavior for the task. (Chinese interface.)

The permission-mode menu in Basic settings; choose the confirmation behavior for the task. (Chinese interface.)

Want to Go Deeper?

See the Advanced Tutorials for the full workflow:

Get Help and Submit Feedback

If you have any questions, bugs, or feature suggestions during configuration or use, please use the official channels listed in Feedback and Suggestions.

Advanced Capability Map

Previous Page

Agent Workspace

Next Page

On this page

What It Can DoQuick StartPermission ModesWant to Go Deeper?Get Help and Submit Feedback