[Feature]: Always show each subagent's model in the Agents panel

[morriisan]

Before submitting

• I searched existing issues and did not find a duplicate.
• I am describing a concrete problem or use case, not just a vague idea.

Area

apps/desktop

Problem or use case

The Agents panel does not consistently show which model each subagent is using.

In the attached screenshot, the agent row shows the agent name, current activity, token usage, and elapsed time, but not the model

Proposed solution

Show the actual model used by each subagent directly in its row in the Agents panel.

For example:

GPT-5.6 Luna -max

Why this matters

Model visibility makes it easier to:

• Confirm that model overrides are working
• Understand differences in speed and output quality
• Debug unexpected subagent behavior
• Compare agents when several are running concurrently

Smallest useful scope

Add a model label to every subagent row, visible while the agent is running and after it completes

Alternatives considered

Currently, the user has to inspect logs, open the child session, or ask the parent agent which model was used. These options are slower, and consume extra tokens. Pain

Risks or tradeoffs

No response

Examples or references

Related model-accuracy reports:

• [Bug]: Sub agents always show Sonnet 5 for model even when explicitly set to other models #6017
• [Bug]: Agents panel mislabels model for depth-2 (nested) agent spawns — shows session default instead of actual serving model #6026

Contribution

• I would be open to helping implement this.

[morriisan]

sol wrote this

---

I traced the flow, and the issue seems to be in the Codex server integration, not the Agents panel.

Codex child agent
→ CodexSessionRuntime
→ CodexAdapter
→ RuntimeSubagent
→ AgentsPanel

The UI already supports displaying model and effort. They are lost earlier because:

• CollabChildAgentState does not store them.
• The synthetic collabAgent/* events do not include them.
• CodexAdapter therefore cannot add them to the resulting task.* events.

A straightforward fix would be to read the model and effort from the completed spawnAgent event, store them on CollabChildAgentState, and pass them through the synthetic events and task.* payloads.

The spawn event may arrive after the child is registered, so the merge should work in either order. thread/settings/updated could also be used later to refine the effective values.

The Agents panel itself should not need changes.

[juliusmarminge]

Thanks for calling out the missing model labels. The orchestrator V2 work has now merged in #2829.

V2 carries subagent model metadata into the agent presentation and displays the reported model with its effort when available. Unknown values remain unknown rather than borrowing the parent's model.

Closing this as delivered. If a specific part is still missing in a build containing V2, please open a focused follow-up with the provider/version and the behavior you are seeing.