fix: translate max_tokens/max_completion_tokens → max_output_tokens in Chat→Responses translator

[Gi99lin]

Problem

When translating from OpenAI Chat Completions format to OpenAI Responses API format (openaiToOpenAIResponsesRequest), the translator passes max_tokens through as-is:

if (root.max_tokens !== undefined) result.max_tokens = root.max_tokens;

The OpenAI Responses API does not accept max_tokens — it requires max_output_tokens. This causes a 400 error for every request routed through OmniRoute to a Responses API endpoint (e.g., Codex):

{"detail":"Unsupported parameter: max_tokens"}

Additionally, max_completion_tokens (the newer Chat Completions field introduced by OpenAI) was not handled at all, meaning clients using the modern field would lose their token limit during translation.

Fix

Replace the direct passthrough with proper field mapping:

-  if (root.max_tokens !== undefined) result.max_tokens = root.max_tokens;
+  // Translate max_tokens / max_completion_tokens → max_output_tokens for Responses API.
+  // The Responses API does not accept max_tokens or max_completion_tokens; it requires
+  // max_output_tokens. max_completion_tokens takes priority as the newer Chat Completions field.
+  if (root.max_completion_tokens !== undefined) {
+    result.max_output_tokens = root.max_completion_tokens;
+  } else if (root.max_tokens !== undefined) {
+    result.max_output_tokens = root.max_tokens;
+  }

• max_completion_tokens → max_output_tokens (preferred, newer field)
• max_tokens → max_output_tokens (fallback for legacy clients)
• max_completion_tokens takes priority when both are present

Reproduction

1. Configure a provider with openai source format targeting an openai-responses upstream (e.g., Codex)
2. Send a Chat Completions request with max_tokens or max_completion_tokens
3. OmniRoute translates and forwards with max_tokens intact → 400 error from upstream

Testing

The existing translator-openai-responses-req.test.mjs test suite should be extended with cases for max_tokens and max_completion_tokens translation.

[gemini-code-assist]

Code Review

This pull request updates the Responses API translation logic to map max_tokens and max_completion_tokens to max_output_tokens, ensuring compatibility with the Responses API while prioritizing the newer max_completion_tokens field. However, the changes introduced several formatting regressions, including broken indentation and unnecessary blank lines across multiple blocks, which need to be corrected to maintain code readability.

[gemini-code-assist]

The indentation of the return statement is broken, and an unnecessary blank line was introduced. This violates standard TypeScript formatting practices and reduces code readability.

Suggested change

	return imgResult;
	return imgResult;

[gemini-code-assist]

The indentation of the error message string is broken, and an unnecessary blank line was introduced. String literals within function calls should maintain proper indentation.

Suggested change

	`Unsupported Responses API feature: tool_choice type '${tcType}' is not supported by omniroute`
	`Unsupported Responses API feature: tool_choice type '${tcType}' is not supported by omniroute`

[gemini-code-assist]

The indentation of the else if block is broken, and an unnecessary blank line was introduced. This makes the control flow harder to follow.

Suggested change

	} else if (contentItem.type === "thinking" || contentItem.type === "redacted_thinking") {
	} else if (contentItem.type === "thinking" || contentItem.type === "redacted_thinking") {

[gemini-code-assist]

The indentation of the assignment is broken, and an unnecessary blank line was introduced. Maintaining consistent indentation is important for code quality.

Suggested change

	result.previous_response_id = root.previous_response_id;
	result.previous_response_id = root.previous_response_id;

[diegosouzapw]

Thanks @Gi99lin for catching this! 🎉 The max_tokens → max_output_tokens translation fix is critical for Responses API routing. We added 3 additional unit tests covering both fields and priority behavior. Integrated into release/v3.6.6 — will ship with the next release!