[FR] FleetView / task-list panel: show which model each agent is running

Resolved 💬 3 comments Opened Jun 12, 2026 by ThatDragonOverThere Closed Jun 16, 2026

Summary

The FleetView multi-agent panel (⏵⏵) shows agent name, runtime, and token count — but not which model each agent is actually running. Please add a model indicator to each agent row.

Current state

◯ general-purpose  Apply tier-1 sweep class fixes     46m 54s · ↓ 539.4k tokens
◯ general-purpose  Fix IV double-÷100 parity bug      30m 29s · ↓ 164.5k tokens
◯ general-purpose  Exit overhaul: backstop + exit      11m 25s · ↓ 219.9k tokens

No model column. Looking at 3 agents running — no way to tell which are Fable-tier vs Sonnet vs Haiku without interrupting each one.

Requested state

◯ general-purpose  [sonnet]  Apply tier-1 sweep class fixes     46m 54s · ↓ 539.4k tokens
◯ general-purpose  [fable]   Fix IV double-÷100 parity bug      30m 29s · ↓ 164.5k tokens
◯ general-purpose  [sonnet]  Exit overhaul: backstop + exit      11m 25s · ↓ 219.9k tokens

Even a short model-tier tag (e.g. , , , ) would be sufficient. Full model ID in a detail view or on hover would be a nice bonus.

Why this matters

1. Cost visibility in autonomous runs. Token count is shown but model tier determines the per-token cost by 5-10×. 539k tokens at Sonnet rates vs Fable rates is a large difference. Right now there is no way to see this inline.

2. Catching silent model switches. There are open bugs for Fable 5 silently routing sub-agents to Fable tier (#67199) and for safety-classifier-triggered model switches (#67937). Neither is catchable in real time from FleetView today because the model isn't surfaced.

3. Verifying agent-spec model assignments. Agent specs can set a model: in YAML frontmatter. FleetView currently gives no confirmation that the spec's declared model is what's actually running — you have to interrupt the agent and ask.

4. Overnight / unattended runs. If a sub-agent silently upgrades or downgrades mid-session, the operator has no way to notice until the run completes and the costs/behavior are reviewed. A model column would make this visible at a glance.

Minimal viable version

A short model tag on each agent row in FleetView. Could be abbreviated (e.g., / / / ) if space is tight. The information is already known to the runtime — this is purely a display gap.

Related

  • #67199 — Fable 5 gate-skip + wrong-tier sub-agent routing
  • #67937 — Fable 5 safety classifier forces mid-session model switch (silent in autonomous context)

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗