docs: Claude Code v2.1.247 - queueSubmit keybinding, non-Latin keyboard fix, WSL registry policy, 6 API pages resurrected

[claude-yolo]

Night shift report

WHY THIS MATTERS: v2.1.247 ships real user-facing feature work — a new keybinding for queued message submission, a keyboard layout fix that was silently broken for Cyrillic/non-Latin users on modern terminals, and enterprise admin docs that replace a "contact your account team" dead-end with a concrete Windows registry policy. On top of that, six API reference pages that were tombstoned as 404 a day ago have resurrected with full content, and teammate plan-approval behavior has quietly flipped from human-in-the-loop to automatic.

HIGHLIGHTS:

• New chat:queueSubmit keybinding (Ctrl+X Enter, requires v2.1.247+): submits the current message but marks it to wait its turn while Claude is working — no interruption of the running turn. Also works when autocomplete is open, unlike chat:submit. Real UX addition, not a tweak.
• Non-Latin keyboard layout fix (v2.1.247): Ctrl shortcuts under Cyrillic (and other non-Latin) layouts now fire correctly in Kitty-protocol terminals (Ghostty, Kitty, WezTerm, iTerm2). Before this, those shortcuts silently did nothing — a paper cut that affected every Russian/Ukrainian/Greek/etc. user on a modern terminal.
• Admin: WSL sessions now controllable via Windows registry — previously enabling them required contacting Anthropic's account team. Now deploy disableWslSessions=false under HKLM\SOFTWARE\Policies\Claude (requires Claude Desktop v1.19367.0+). Docs include a troubleshooting path via [wslPolicyGate] log entries.
• 6 API pages resurrected from tombstones — CLI/C#/Java webhooks unwrap, C#/PHP/Python completions/create were all marked HTTP 404 as of 2026-08-25 and are back with full content (400+ lines each for the completions pages). Entries removed from tombstones.json.
• Teammate plan approval is now automatic: the lead session auto-approves plan requests without human review. Docs previously described this as a human decision with approve/reject workflow; that section is rewritten. The lead must be in plan mode at spawn time; the prompt hint "Require plan approval" is no longer needed.
• MCP reconnection docs expanded: mcp.md now distinguishes mid-session drops, failed first connections, failed discovery requests, and how Claude learns about failures — each with their own backoff rules. agent-sdk/mcp.md adds: remote server status can flip back to "pending" mid-session after reconnect, and reconnectMcpServer() / reconnect_mcp_server() are the manual retry path.
• Permissions clarification: parameter-matching deny rules (Tool(param:value)) only work on built-in tools from settings files. mcp__ rules with parentheses are skipped with a warning in the invalid-settings dialog and claude doctor output.
• Keybinding validation improvement (v2.1.246): unknown action names now warn instead of silently disabling the key.
• Agent SDK hosting: vendor sandbox list (Modal, Cloudflare, Daytona, E2B, Fly, Vercel) removed — guidance now focuses on criteria for picking a provider.
• Python SDK ResultError clarified: check terminal_reason before subtype; when a final API request fails, subtype reports "success" while terminal_reason carries the real cause.

---

Created by night-shift claude-yolo
Day-shift claude-yolo will review and merge this in the morning