[Bug]: Claude Opus 4.7 default effort `xhigh` crashes Claude CLI on session start

[wopas]

Before submitting

• I searched existing issues and did not find a duplicate.
• I included enough detail to reproduce or investigate the problem.

Area

apps/server

Steps to reproduce

1. Build T3 Code from current main (post-feat: add Claude Opus 4.7 to built-in models #2072) or use any build that includes Claude Opus 4.7
2. Ensure Claude CLI 2.1.111 (the gated minimum) is installed and on PATH
3. Select "Claude Opus 4.7" in the model picker
4. Leave effort at the default (xhigh)
5. Send any message

Result: the Claude CLI subprocess exits with code 1 before the turn starts. T3 Code surfaces this as:

ProviderAdapterRequestError: Provider adapter request failed (claudeAgent) for turn/setPermissionMode: Claude Code process exited with code 1

Workaround: after selecting Opus 4.7, manually change effort from xhigh to any other value (high, max, medium, low all work).

Likely cause (guess): apps/server/src/provider/Layers/ClaudeProvider.ts:55 sets xhigh as isDefault: true for claude-opus-4-7. Claude CLI 2.1.111 doesn't appear to accept xhigh as a valid effort value, so the subprocess dies at spawn.

Environment:

• T3 Code: built from main @ 19d4740 (since feat: add Claude Opus 4.7 to built-in models #2072 isn't in any release yet)
• OS: Windows 11 Pro (26200)
• Claude CLI: 2.1.111

Related: #2072 (added 4.7), #2075 (feature request, separate from this bug)

Expected behavior

Selecting Claude Opus 4.7 and sending a message with the default effort should work without crashing the Claude CLI subprocess. Either the default effort should be a value the gated-minimum CLI accepts, or unsupported effort values should be rejected/normalized before the spawn.

Actual behavior

The Claude CLI subprocess exits with code 1 at session startup before the turn can begin. T3 Code surfaces this as a "Provider turn start failed" runtime error in the work log (full stack trace in the Logs section below).

Impact

Minor bug or occasional failure

Version or commit

Built from main @ commit 19d4740 (local build, since #2072 isn't in any release yet)

Environment

• OS: Windows 11 Pro (build 26200) - Claude CLI: 2.1.111 (the gated minimum in MINIMUM_CLAUDE_OPUS_4_7_VERSION) - Node/Bun: ran build via bun 1.3.11

Logs or stack traces

ProviderAdapterRequestError: Provider adapter request failed (claudeAgent) for turn/setPermissionMode: Claude Code process exited with code 1
    at toRequestError (file:///C:/Users/fn101/AppData/Local/Programs/t3-code-desktop/resources/app.asar/apps/server/dist/bin.mjs:13398:9)
    at catch (file:///C:/Users/fn101/AppData/Local/Programs/t3-code-desktop/resources/app.asar/apps/server/dist/bin.mjs:14862:22)
    ... (truncated)
  [cause]: Error: Claude Code process exited with code 1
    at QX.getProcessExitError (...@anthropic-ai/claude-agent-sdk/sdk.mjs:58:8018)
    at ChildProcess.J (...@anthropic-ai/claude-agent-sdk/sdk.mjs:58:11026)
    at Object.onceWrapper (node:events:623:26)
    at ChildProcess.emit (node:events:520:35)
    at ChildProcess._handle.onexit (node:internal/child_process:294:12)

Screenshots, recordings, or supporting files

No response

Workaround

After selecting Opus 4.7 in the model picker, manually change effort from xhigh to any other value (high, max, medium, low all work).

[juliusmarminge]

Should hopefully be fixed by #1729 which will be included in the next release