C:/Program Files/Git/schedule fails with "trouble connecting with your remote claude.ai account" — persistent, no public status incident

Summary

/schedule (the schedule skill) has been failing persistently for ~2 hours with the same error message, with no corresponding incident on https://status.claude.com.

Error string: We're having trouble connecting with your remote claude.ai account to set up a scheduled task. Please try /schedule again in a few minutes.

First failure: 2026-05-31 ~16:48 UTC (Windows local; UTC offset 0 if BST applied).
Latest failure: ~19:00 UTC (8 attempts total, each spaced ~5 min apart plus one final attempt after a full app restart).

Mitigations tried (none worked)

1. Waited 5 minutes between attempts — same error every time.
2. Signed out of claude.ai in a browser, signed back in, retried — same error.
3. Fully quit and relaunched the Claude desktop app (system tray → Quit, then reopen) — same error.

Client-side evidence

• Desktop app main.log shows the scheduled-tasks MCP server registered cleanly with every session (serverCount=2, 9 total servers). No registration or handshake failures.
• OAuth tokens for api.anthropic.com are healthy: [oauth] looking up token / [oauth] using cached token succeeding for scopes user:inference user:file_upload user:profile.
• claude.ai-web.log shows TypeError: Failed to fetch from cu.fetch in vendor-CaKhGNTw.js at 18:30:20 UTC today. Same pattern recurs at similar times on prior days: 2026-05-29 16:42 / 19:39 / 20:32 / 21:15; 2026-05-30 13:47 / 16:47; 2026-05-31 12:01 / 13:19 / 18:30. Suggests a recurring webview-side fetch failure on claude.ai endpoints, not just today.
• No \"trouble connecting\" string or stack trace surfaced into the desktop main.log — the message comes from the schedule skill's error path on the webview side.

Network probes from the same host

• ping claude.ai → normal RTT (~100ms).
• curl https://claude.ai/ → HTTP 403 (Cloudflare default for unauth).
• curl https://claude.ai/api/scheduled_tasks → returns Cloudflare \"Just a moment…\" challenge HTML (HTTP 403, cType: managed).
• curl https://api.anthropic.com/ → HTTP 404 (expected).
• curl https://assets-proxy.anthropic.com/... → HTTP 200.
• curl https://mcp-proxy.anthropic.com/ → HTTP 404 (expected).
• https://status.claude.com/api/v2/summary.json: \"indicator\": \"none\", \"description\": \"All Systems Operational\", no unresolved incidents, all six listed components (claude.ai, Console, API, Claude Code, Cowork, Claude for Government) operational. Scheduled Tasks is not a separately listed component.

Environment

• App: Claude (desktop) 1.7196.3.0, Windows.
• Org ID: bfc14fb8-88f3-4818-bc89-bac0196e1f45.
• Skill version: schedule (under local-agent-mode-sessions/skills-plugin/.../skills/schedule/SKILL.md).

Hypothesis

Either (a) the Scheduled Tasks back-end is rejecting requests in a way that isn't reflected on the public status page, or (b) Cloudflare in front of claude.ai/api/scheduled_tasks is challenge-bouncing the embedded Chromium webview's fetch even with a freshly signed-in session and a freshly restarted webview process. Other claude.ai webview features work normally on this host during the same window.

Asks

1. Confirm whether there's a back-end issue affecting create_scheduled_task that isn't on the public status page.
2. If client-side, suggest a webview state clear short of reinstalling the app.
3. Either way: surface a more specific error string from the back-end response into the skill output, so the next user with this can distinguish auth / Cloudflare / 5xx / quota.

Happy to share any specific log lines or a HAR if useful.