`/ultraplan` remote session never reaches ExitPlanMode — two consecutive 90-min timeouts, remote container appears to never start

Resolved 💬 3 comments Opened Apr 13, 2026 by inivosdigital Closed Apr 17, 2026

Summary

Running /ultraplan from the Claude Code CLI spawns a remote web session that never makes progress. The local CLI prints the claude.ai/code/session_... link, but after 90 minutes the harness times out with: "ExitPlanMode never reached after 90 minutes (the remote container failed to start, or session ID mismatch?)". Reproduced twice back-to-back with different session IDs, so it's not a transient glitch on a single session.

Environment

  • Claude Code CLI running on linux/arm64 (Orange Pi 5 Plus, Debian-based, Linux 5.10.110-rockchip-rk3588)
  • Model: claude-opus-4-6[1m]
  • Shell: bash
  • Interactive session (not a subagent)

Repro

  1. Invoke /ultraplan with a long-form planning prompt (~8KB of text, deep architectural design request — nothing exotic, just a detailed brief).
  2. CLI prints: ◇ ultraplan · Monitor progress in Claude Code on the web https://claude.ai/code/session_<id>
  3. Wait.
  4. 90 minutes later: Ultraplan terminated: ExitPlanMode never reached after 90 minutes (the remote container failed to start, or session ID mismatch?)

Observed

  • Happened on two separate invocations in the same local session:
  • session_01YReV3JRqULs1xL7SXbTSu4
  • session_01JAsDmJ6ddrcBMo9q1yYotT (task id rgligwvsh, spawned ~2026-04-13)
  • Both failed identically.
  • Local remote-agents/remote-agent-<id>.meta.json was written successfully, so the CLI side of the handshake worked; the remote side never progressed to ExitPlanMode.

Expected

Remote container spins up, runs the plan, ExitPlanMode fires, results stream back to the CLI.

Notes

  • Recent changelog mentions: "/ultraplan and other remote-session features now auto-create a default cloud environment instead of requiring web setup first" — possible that this auto-create path is silently failing for my account/org but the CLI isn't surfacing the error.
  • Would be very helpful if, on timeout, the CLI could report the remote container's last-known state (queued / starting / crashed / never-created) rather than just "never reached ExitPlanMode."

View original on GitHub ↗

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