Repository navigation
Feature request: Show subagent tool call details #1322
Description
Summary
When subagents run, the CLI shows minimal information about what they're doing. In contrast, Copilot Chat in VS Code allows you to see and drill into the tool calls made by subagents.
Current behavior
The subagent details panel shows:
- Status (e.g., "Completed (24s)")
- ID, Type, Description
- First 10 lines of Prompt
- First 10 lines of Result
However, there's no visibility into:
- Which tools the subagent called
- Tool parameters and responses
- Intermediate steps/reasoning
Expected behavior
Add the ability to view subagent tool calls, similar to how Copilot Chat displays them. This would help users:
- Understand what the subagent actually did
- Debug issues when subagents produce unexpected results
- Learn from the subagent's approach to tasks
Additional context
Screenshot showing the current limited subagent details view:
The CLI already surfaces tool calls for the main agent - extending this to subagents would provide consistency and better observability.
Reacted by mokulus-btp, skurzbac, fluxys-indeherb, tore-unumed, Jatin Sharma, Adam Kecskes, AlexBergzeit, manfreed, Hallur Holm Johannessen, Jatin Sharma and 15 more
Reacted by Sebastian Carlos
Activity
cephalin commentedon Feb 6, 2026
Here's how the custom agent runs as a subagent run looks in Copilot CLI:
and here it is when the same custom agent runs as a subagent in Copilot Chat:
indeherb commentedon Mar 23, 2026
+1 on this. Running multi-agent workflows on hard problems (3-5 background agents in parallel, 10-30 min runtimes), the black-box experience is painful.
The main agent's reasoning is visible via ctrl+t, but sub-agents give you nothing until they complete or timeout. For long-running agents, you want to know if they're on track or burning tokens on a dead end.
Beyond tool call visibility (which would already help a lot), live access to the reasoning/thinking stream would be the real game changer. Something like
/tasks --watch <agent_id>showing a periodically-refreshed reasoning stream.Current workaround: instructing agents to write intermediate progress to disk files — effectively a DIY thought log. Works, but it'd be great to have this natively.
Rogn commentedon Apr 30, 2026
I ran into the same thing and ended up building a small external observer for this:
https://github.com/Rogn/copilot-cli-agent-observer
It’s not a replacement for having this built directly into Copilot CLI, but it may be useful while this is still missing. It opens a read-only observer window that shows the agent flow, including subagents, tool calls, messages and timings.
A couple of caveats: it requires Copilot CLI experimental mode, and it’s still early. I mainly built it because I wanted more visibility into exactly the problem described here.
cjoprey commentedon May 6, 2026
+1 to this - in OpenCode I can click on the message showing a subagent started and view the full output in the same ways as I can for the main agent, including all reasoning and tool calls. It's one of the most painful missing features in Copilot CLI for me,
miguelcosta30 commentedon May 19, 2026
+1, tbh the current way kinda sucks
Smaswegen commentedon May 19, 2026
``
copilot-cli-bot commentedon Sep 16, 2026
This appears to be resolved as of GitHub Copilot CLI v1.0.85. Please update to that release or a newer stable version. We're closing this as completed. If it still happens, please comment with your version and reproduction steps.