[FEATURE] Desktop v2: show effective model and reasoning effort for running subagents #58809

Summary

Show the actual effective model and reasoning effort for each child session directly in the running-subagents list in OpenCode v2 Desktop.

When a parent delegates work to multiple subagents, users should be able to see which model and reasoning level each child is actually using without opening every child session or inspecting configuration files.

Current issue

The Desktop running-subagents list does not provide the per-child effective model and reasoning effort needed to understand a delegated run at a glance. A child can use a different model or effort level from the parent or from another child.

Displaying the parent's current selection, an agent's configured defaults, or assumed inherited settings is insufficient: these can differ from the values resolved and persisted for the child's actual execution. Configuration alone does not establish what a particular child run used.

This makes it difficult to verify delegation choices, compare running children, investigate unexpected behavior, or understand the cost and latency implications of a run.

Expected behavior

Each child row in the Desktop running-subagents list should display:

• The child agent name or task label.
• The effective model used by that child.
• The effective reasoning effort/variant used by that child.
• Its execution status, such as Running or Completed.

For example:

• Explore · Qwen Flash · low · Running
• Review · DeepSeek Flash · high · Running
• Build · GPT-6 · medium · Completed

These are illustrative display examples; model names and supported effort values depend on the provider.

The information should be visible from the parent session's running-subagents list. Users should not have to navigate into each child to discover it. Retain this information for completed children wherever their rows remain visible.

Technical requirements

1. Use the child's persisted/effective execution data as the source of truth. Resolve the provider/model identity and reasoning effort/variant for that specific child execution after applicable defaults, inheritance, and overrides have been applied.
2. Do not derive the displayed values solely from the parent's current model/variant selector, static agent configuration, or assumed inheritance. Changing the parent's settings later must not relabel an existing child run.
3. Associate metadata with the correct child session/run so concurrent children using different models or effort levels cannot be confused.
4. If the required execution metadata is not currently persisted or exposed to Desktop, persist/expose it through the appropriate session/run data and updates. The requested behavior is end-to-end, rather than a UI label populated with configured defaults.
5. Update the row when effective execution metadata becomes available, and preserve the recorded values across refreshes, reconnection, and reopening the session.
6. If a child changes model or effort during its session, show the effective values for the currently running execution; once completed, preserve the values for its latest execution. A tooltip or expanded detail can explain changes if needed.
7. Handle unsupported reasoning effort, provider defaults, and missing metadata explicitly. Use an accurate state such as "Not supported", "Provider default" (only when known), or "Unknown"; do not invent a low/medium/high value.
8. Use a readable model display name in the row, with full provider/model identifiers and the recorded effort/variant available in a tooltip or expanded detail where needed.
9. Keep the list compact and accessible, including clear status text and sensible truncation for long names.

Reproduction steps

1. Open OpenCode v2 Desktop and start a parent session.
2. Delegate tasks to multiple child agents, using different supported models and reasoning effort/variant values for different children.
3. While the children are running, inspect the running-subagents list from the parent session.
4. Try to determine the actual model and reasoning effort for each child from that list.
5. Compare the visible information with each child's recorded execution metadata; also check after a child completes and after reopening the parent session.

Current result: the list does not expose the required per-child effective model and reasoning effort clearly enough to verify the running children at a glance.

Expected result: each row shows the child's actual persisted/effective model, effort, and status, including when they differ from the parent's current selection or configured defaults.

Scope and acceptance criteria

• This request specifically targets the running-subagents list in OpenCode v2 Desktop.
• Each child independently shows its actual effective model and effort.
• Different concurrent children can display different model/effort combinations.
• Parent setting changes do not overwrite historical child execution metadata.
• Running and completed rows remain accurate after refresh/reopening.
• Missing or unsupported effort is represented honestly.

Related Issues

• [FEATURE]:Show reasoning/variant level for subagents in the UI #26266 — Original request to display subagent reasoning/variant levels; closed automatically.
• ui: show active subagent model and reasoning effort in bottom bar #57366 — Similar v2 request; closed as a duplicate.

Both references above are closed. This request makes the Desktop running-subagents-list scope and the requirement for actual persisted/effective per-child execution values explicit.