[Feature]: Agents panel should also show CLI-delegated child agents (codex exec / opencode run / etc.), not only provider-native subagents

[JorgeMenaDev]

Before submitting

• I searched existing issues and did not find a duplicate. ([Bug]: Desktop/web thread shows no sign of delegated CLI-agent work: silent while it runs, idle-looking when backgrounded, spontaneous wake after #5043 reports the same underlying gap as a pre-Agents-panel bug on the thread surface; this is a feature request scoped to the new Agents panel shipped in PR feat: native subagent & workflow observability #5219.)
• I included enough detail to reproduce or investigate.

Area

apps/web

Feature request

Render CLI-delegated child agents in the Agents panel, not only provider-native subagents.

PR #5219 ("native subagent & workflow observability", v0.0.32-nightly.20260806.1012) is a great step: Claude Code Task-tool spawns and Codex collab agents now show up in the Agents panel with model, tokens, tool count, and status. Thank you for this.

But a very common delegation shape is still invisible there: the primary agent spawning another vendor's agent CLI as a child process — e.g. a Claude thread running codex exec "…" or opencode run "…" --auto --format json through its Bash tool (foreground or backgrounded). Multi-vendor routing like this is how many of us pick a cheap/fast model per task while keeping one conversation.

Repro (today, on v0.0.32-nightly.20260806.1012):

1. In a Claude thread, spawn a native subagent via the Task tool → it appears in the Agents panel with full attribution. ✅
2. In the same thread, ask the agent to run opencode run "say hi" --auto --format json or codex exec "say hi" as a background Bash command → the child agent runs to completion but never appears in the Agents panel; the only trace is the generic running-task line in the transcript. ❌

Both are "my thread is running a fleet," but only one is observable in the panel.

Suggested directions (either would help):

• Heuristic attribution: recognize well-known agent CLIs (codex exec, opencode run, claude -p, grok, cursor-agent, …) in Bash/exec tool events and render them as panel members with elapsed time and status derived from the process/tool lifecycle.
• Explicit annotation: a documented way for the primary agent (or a wrapper script) to declare "this command is a child agent: vendor, model, task label" so the panel can render it first-class — this seems to fit the Orchestrator v2 schema alignment mentioned in feat: native subagent & workflow observability #5219.

Related: #5043 (same gap reported on the thread/transcript surface before the Agents panel existed).

[SpaceM88]

Thanks for opening this. I have a related observability request for the Agents panel.

For every visible subagent, could the panel also show runtime attribution metadata when available?

• Provider/runtime: Codex, Claude, OpenCode, etc.
• Model ID
• Reasoning effort/level, such as low, medium, high, or extra-high
• Status, elapsed time, and token usage

In my current view, an agent can appear as running without making it clear which model or reasoning configuration is actually being used. That makes it difficult to understand cost, latency, capability differences, and whether the requested configuration was applied.

I am not asking for private chain-of-thought or raw internal reasoning. I only mean the configured/runtime metadata. A compact secondary line or expandable details view would be enough. If a provider does not expose a field, displaying "Unknown" or "Not reported" would be helpful.

This would make the Agents panel much more useful for monitoring and debugging multi-agent workflows.