[Feature]: ACP agent does not show real-time progress/tool calls in TUI

Author: longlifeCreated Sep 17, 2026Updated Sep 17, 2026
Labelsenhancement

Description

ACP agent does not show real-time progress/tool calls in TUI

Description

I'm using the acpAgents feature to delegate implementation work to Cursor Agent via ACP.

My setup works correctly: OMO-Slim successfully starts cursor-agent acp, delegates the task, and receives the final result.

However, while Cursor Agent is actually working, the OMO-Slim TUI provides almost no visible progress. I basically have to wait until Cursor finishes before I can see the result.

Configuration

json
{
  "acpAgents": {
    "cursor-fixer": {
      "command": "C:\\Users\\Administrator\\AppData\\Local\\cursor-agent\\agent.cmd",
      "args": ["acp"],
      "description": "Cursor ACP agent for implementation and UI/design code work.",
      "orchestratorPrompt": "@cursor-fixer\n- Role: Primary agent for implementation AND UI/design code work\n- Prefer this over @fixer and @designer for all code changes and UI implementation",
      "permissionMode": "allow",
      "timeoutMs": 1800000
    }
  }
}

Current behavior

The delegation works, but during a long-running Cursor task the OMO-Slim UI appears to be waiting:

orchestrator
  └─ @cursor-fixer
       └─ cursor-agent acp
             ...
             ...
             ...
       [long period with little/no visible activity]
             ...
       final result

For example, Cursor may be reading files, searching the codebase, editing files, running tests, etc., but these activities are not visible in the parent OMO-Slim TUI.

This makes it difficult to tell whether the ACP agent is:

  • actively working
  • waiting on a tool
  • running tests
  • stuck
  • or has actually stopped

Expected behavior

It would be very useful if ACP session updates from the external agent could be surfaced in the OMO-Slim TUI in real time.

For example:

@cursor-fixer

▸ Reading src/server.js
▸ Searching for TravelApply
▸ Editing public/js/cards.js
▸ Running npm test
✓ 12 tests passed
▸ Editing test/card-action.test.js

I'm not asking for the model's private chain-of-thought.

What would be useful is the normal agent execution/progress information exposed by ACP, such as:

  • tool_call
  • tool_call_update
  • plan
  • agent_message_chunk
  • tool status / completion
  • possibly file locations associated with tool calls

Question

Does the current acpAgents implementation intentionally only surface the final/agent-message output from session/update, rather than forwarding the external ACP agent's tool/progress events to the OMO-Slim TUI?

If so, would it be possible to expose these ACP events as a live child-agent session in the TUI?

This would make external ACP agents such as Cursor Agent much easier to monitor during long-running coding tasks.

Thanks!

Extra context

No response

Source: alvinunreal/oh-my-opencode-slim