/remote-control fails with 'remote credentials fetch failed' inside a long-lived session; standalone claude remote-control works

Summary

/remote-control run inside an already-running interactive session fails immediately with remote credentials fetch failed, while the standalone claude remote-control from a fresh process in the same project succeeds (reaches the spawn-mode prompt).

Environment

• Claude Code v2.1.212
• macOS (Darwin 25.5.0), zsh, terminal CLI
• Auth: claude.ai OAuth login, apiProvider: firstParty, subscription max (no ANTHROPIC_API_KEY in the environment)

Repro

1. Start an interactive claude session; use it normally for a few hours (long-lived session).
2. Run /remote-control in the session → fails instantly with remote credentials fetch failed. Reproduced 3 times over ~1 hour in the same session.
3. In a separate terminal, same directory: claude remote-control → works; prints the spawn-mode chooser ([1] same-dir / [2] worktree), no credential error.

Diagnostics already ruled out

• claude auth status: loggedIn: true, authMethod: claude.ai, firstParty, subscriptionType: max
• curl -I https://api.anthropic.com → responds in ~120 ms (connectivity fine)
• Version is past the v2.1.206 error-message improvements

Hypothesis

The in-session slash-command path mints the remote-control credential with the session's in-memory OAuth token; in a long-lived session that token appears stale where a fresh process refreshes before the fetch. Only the in-session path fails.