[Fix] Kimi Code sends incorrect reasoning parameters for subscription models

[zoomote]

​Opened on behalf of @taltas. Follow up by mentioning @roomote, in the web UI, or in Discord.

Related GitHub Issue

Related: #1316

Follow-up to: #1325

Description

Kimi Code now applies protocol-specific request behavior for subscription and API-key models. K3 keeps the direct API's 131K output default, uses low/high/max reasoning effort with high as the default, and omits temperature. K2.7 Code sends preserved thinking instead of reasoning_effort and carries reasoning context through tool-result continuations.

The generic OpenAI-compatible handler exposes narrow request and message-conversion hooks so Kimi Code can customize these fields without duplicating the full request pipeline. Discovered server capacity still overrides local fallback metadata without guessing a protocol for unknown future model IDs.

Test Procedure

Run pnpm test and pnpm check-types from the repository root. The provider and fetcher suites include wire-level coverage for all four model IDs, temperature omission, custom effort, OAuth retry, server metadata overrides, and preserved thinking across a tool-result plus trailing-text continuation.

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue (see "Related GitHub Issue" above).
• Scope: My changes are focused on the linked issue (one major feature/fix per PR).
• Self-Review: I have performed a thorough self-review of my code.
• Testing: New and/or updated tests have been added to cover my changes (if applicable).
• Visual Snapshot (UI changes only): Not applicable; this changes provider request behavior without changing extension UI.
• Documentation Impact: Documentation is updated in Zoo-Code-Docs PR [Fix] Windows CI tests flake and unit-test cache stays cold #43.
• Contribution Guidelines: I have read and agree to the Contributor Guidelines.

Visual Snapshots

Not applicable.

Videos (interaction / animation only)

Not applicable.

Documentation Updates

See Zoo-Code-Org/Zoo-Code-Docs#43.

Additional Notes

PR #1325 correctly fixes Kimi K3 through OpenRouter, whose 32,768-token upstream cap differs from the direct Kimi Code API. This PR intentionally keeps the direct provider's 131,072-token output default.

Related PRs

• Zoo-Code-Docs: [Docs] Clarify Kimi Code model reasoning behavior Zoo-Code-Docs#43

Get in Touch

@taltas

[codecov]

Codecov Report

❌ Patch coverage is 91.66667% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/api/providers/kimi-code.ts	83.33%	0 Missing and 2 partials ⚠️

📢 Thoughts on this report? Let us know!