Remote control: dead environment (404) shows misleading "Sign in again to verify your device" modal instead of environment-gone state

Environment

• Claude Desktop app 1.32352.1, macOS 26.5 (Apple Silicon)
• Claude Code CLI 2.1.233 running claude remote-control --spawn worktree under a LaunchAgent (KeepAlive=true)

What happened

1. The remote-control host serving environment env_01LzXh… died with a bare Error: Not found after the backend had stopped recognizing its state (it had earlier logged Error: CCR v2 worker registration failed for session cse_01H4…: Request failed with status code 404 twice — for a session that had already reported "Session completed" the previous day).
2. The LaunchAgent relaunched the host 2 seconds later. It could not resume the preserved environment and silently registered a fresh environment (startup banner switched from "Continue coding…" to "Code anywhere…"). No warning that the old environment and its chats were being abandoned.
3. Every chat bound to the old environment now 404s when opened. From the desktop webview log:

[REACT_QUERY_CLIENT] QueryClient error: {"type":"not_found_error","statusCode":404,
 "endpoint":"/v1/environment_providers/private/organizations/<org>/environments/env_01LzXh…",
 "method":"GET", "rawBody":"…Environment env_01LzXh… not found."}

(e.g. request_id req_011CeBdzpqK7pSHhMHmahbDq, 2026-08-19 ~06:32 UTC)

The bug: misleading sign-in modal

Opening such a chat pops a modal — "Sign in again to verify your device — Please sign in again to continue your session" (Not now / Sign in) — plus a toast "Remote connection setup error". But auth is completely healthy: the 404s above are authenticated responses, and there is not a single 401/403 in the logs for the whole incident window. Signing in again changes nothing, so the user repeatedly chases sign-in/keychain/auth fixes for what is actually an orphaned-environment state. This recurs on every click of an affected chat.

The accurate banner ("Remote environment unavailable — The remote environment this session was connected to is no longer available. Start a new session to continue.") also renders in the same view, so two contradictory diagnoses are shown at once, and the sign-in modal is the one that takes focus.

Expected

• No sign-in/verify-device modal when the failure is an environment 404 with valid auth — show only the environment-unavailable state.
• Ideally a terminal state for chats whose environment is gone, and/or a way to re-bind them to a live environment on the same machine (machine identity is stable via remoteControlMachineId).
• CLI side: after a crash, claude remote-control should say loudly "could not resume environment env_… (not found), registering a new one — existing chats will be orphaned" instead of silently minting a fresh environment. (This has happened twice on this machine — the earlier occurrence followed a DNS outage: "Persistent errors for 10 minutes, giving up.")

Happy to provide fuller logs (main.log / claude.ai-web.log / remote-control host log) on request.

🤖 Diagnosed with Claude Code