[BUG] Agent view: dispatched task intermittently never starts — spare worker not promoted to fleet (job state "working" but daemon roster still "spare")

Preflight Checklist

• [x] Searched existing issues — not reported
• [x] Single bug report
• [x] Latest version

What's Wrong?

In the interactive agent view, typing a prompt into the bottom input and pressing Enter once intermittently creates a task that never starts. The row shows the staged prompt but no model turn ever runs. Manually opening the task and re-submitting starts it every time.

I captured the daemon state during a live stall. Root cause is a warm-spare handoff race:

• The background daemon pre-spawns idle spare workers. On dispatch, a spare should be promoted spare→fleet, seeded with the prompt, and triggered to take its first turn.
• For the stalled task, the job state flipped to working with the prompt recorded (firstTerminalAt: null)…
• …but the daemon roster for that same worker still shows source: "spare", seed.intent: "" — never promoted, never seeded.
• daemon.log shows bg spawned <id> (spare) with no subsequent (fleet) line (workers that actually run are re-logged as fleet).
• No session transcript file is created; the worker process stays alive but idle (low CPU).

So the job/UI side and the daemon roster disagree about the same worker — UI thinks it is "working" with my prompt, daemon thinks it is an idle spare with no intent. Re-sending the prompt manually pushes it down the pty and wakes the idle spare. Because it depends on the timing of spare-consumption vs. promotion, it is intermittent — not correlated with prompt length.

Evidence (redacted — auth tokens / pipe nonces removed):

daemon.log:
  [09:17:09] [bg] bg spawned 90605c6f (spare)     <- pre-warmed spare
  [09:17:48] [bg] bg spawned 5d12001b (spare)     <- replacement spare (not the stall)
  (no "(fleet)" line ever appears for 90605c6f -> never promoted)

jobs/90605c6f/state.json:   { "state": "working", "intent": "<my prompt>", "firstTerminalAt": null }
daemon/roster.json[90605c6f]: { "source": "spare", "seed": { "intent": "" } }   <- mismatch
session transcript for 90605c6f: does not exist

<img width="3006" height="179" alt="Image" src="https://github.com/user-attachments/assets/7263b5a7-9322-4cb8-a066-d1b641837391" />

<br/>

<img width="3054" height="634" alt="Image" src="https://github.com/user-attachments/assets/1bd2bf84-80c0-4650-9053-0640b0a3e7b7" />

What Should Happen?

Per the docs, one Enter starts the session immediately. The bound spare should be promoted to fleet, seeded with the prompt, and started — and the daemon roster should agree with the job state.

Steps to Reproduce

1. Run claude with agent view (background daemon with a warm-spare pool active).
2. Type a prompt in the bottom input, press Enter once.
3. Intermittently: the task row appears with the staged prompt but never runs. Job state = working while the daemon roster keeps the worker as an idle spare; no (fleet) promotion is logged and no transcript is written. Manually re-submitting starts it.

Error Messages/Logs

None surfaced to the user — the failure is silent (no (fleet) promotion, no transcript, no error).

Claude Model

Any (default Opus 4.8)

Is this a regression?

Yes — last working version unknown.

Last Working Version

No response

Claude Code Version

2.1.179 (Claude Code)

Platform

Claude subscription

Operating System

Windows 11

Terminal/Shell

Windows Terminal (Command Prompt / cmd.exe)