fix: accept reasoning effort none when translating to Messages

[alongubkin]

Fixes ALIEN-1123.

Problem

The gateway translates OpenAI requests (Chat Completions or Responses) to Anthropic Messages for Claude models served that way, such as Bedrock Claude. The translator rejects any field Messages can't represent, and reasoning_effort was on that list unconditionally. So a request with reasoning_effort: "none" (Responses: reasoning: { effort: "none" }) failed with a 400, and some OpenAI clients send that value by default.

Fix

"none" asks for no reasoning, which is what a Messages request without thinking already does, so it's representable without loss. The translator now accepts reasoning_effort when it's null or "none" and drops it, the same way it already handles default-valued store, background and truncation. Any other effort (low, high, …) is still rejected, since mapping those to thinking budgets is a separate decision.

Tests

• New reasoning_effort_none_needs_no_reasoning_in_messages: a Responses request with effort none translates to Messages with no reasoning field; effort high is still rejected.
• cargo test -p alien-ai-gateway passes.

🤖 Generated with Claude Code

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Adjusts protocol translation for reasoning effort parameter.

The PR appears safe to merge; no actionable regression was identified.

Summary

The PR allows reasoning_effort values of null or "none" when translating to Messages, removes the field from the outgoing request, and retains rejection of other effort values. The new test covers the Responses translation path and rejection of "high".

_{Reviews (1) · Last reviewed commit: "fix: accept reasoning effort none when t..."}