fix: strip Claude title reasoning effort

[thomasmaerz]

Summary

• Route all outgoing JSON body rewrites through a shared request payload sanitizer.
• Strip reasoning_effort/reasoningEffort only for Claude requests that match OpenCode's hidden title-generation prompt, avoiding Anthropic OAuth's temperature/thinking validation error.
• Preserve reasoning options for normal Claude chat requests and keep existing Gemini schema sanitization behavior.

Fixes #30.

Testing

• npm test
• npm run check:exports

[gemini-code-assist]

Code Review

This pull request refactors the request payload sanitization logic to support stripping reasoning_effort and reasoningEffort from Claude model title generation requests, while keeping the existing Gemini tool schema sanitization. It introduces several helper functions to identify Claude models and detect title prompts, along with corresponding unit tests. Feedback on the changes highlights two issues: first, the use of logical OR assignment (||) for tracking changes is fragile and prone to short-circuiting bugs if refactored; second, in isOpenCodeTitlePrompt, if payload.messages is present but does not contain the title prompt, the function immediately returns false, skipping the evaluation of payload.input.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

Updating the changed flag using logical OR (||) with function calls on the left-hand side is correct but highly fragile and non-idiomatic. If a future refactoring reorders the operands (e.g., changed = changed || ...), the function call will be short-circuited and skipped when changed is already true. It is much safer and more readable to evaluate the functions independently first.

Suggested change

	let changed = false;
	changed = stripClaudeTitleReasoningEffort(clonedPayload) || changed;
	changed = sanitizeGeminiToolSchemas(clonedPayload) || changed;
	return changed ? JSON.stringify(clonedPayload) : undefined;
	const changedClaude = stripClaudeTitleReasoningEffort(clonedPayload);
	const changedGemini = sanitizeGeminiToolSchemas(clonedPayload);
	const changed = changedClaude || changedGemini;
	return changed ? JSON.stringify(clonedPayload) : undefined;

[gemini-code-assist]

If payload.messages is an array, the function immediately returns the result of messages.some(...). If it does not match, the function returns false and completely skips checking payload.input (even if payload.input is present and contains the title prompt). It is more robust to only return true if a match is found, and otherwise continue checking the remaining fields.

function isOpenCodeTitlePrompt(payload: Record<string, unknown>): boolean {
  if (contentContainsTitlePrompt(payload.instructions)) return true;

  const messages = payload.messages;
  if (Array.isArray(messages)) {
    const hasPrompt = messages.some((message) => {
      if (!isRecord(message) || message.role !== 'system') return false;
      return contentContainsTitlePrompt(message.content);
    });
    if (hasPrompt) return true;
  }

  const input = payload.input;
  if (Array.isArray(input)) {
    const hasPrompt = input.some((item) => {
      if (!isRecord(item) || item.role !== 'system') return false;
      return contentContainsTitlePrompt(item.content);
    });
    if (hasPrompt) return true;
  }

  return false;
}