fix(web): submit thinking level verbatim and drop the hardcoded default by wbxl2000 · Pull Request #1673 · MoonshotAI/kimi-code

wbxl2000 · 2026-07-14T06:01:56Z

Related Issue

None — the problem is explained below.

Problem

kimi-web silently rewrote the user's thinking level before it reached the backend, diverging from the CLI (TUI):

Every prompt path (prompt submit, steer, skill activation, BTW side chat) coerced the stored level onto the target model's declared support_efforts — e.g. a carried-over max was downgraded to the model's default high, so what ran did not match what the user picked.
With nothing persisted, the web defaulted to a hardcoded high, a frontend invention with no CLI counterpart. The CLI sends no thinking override when [thinking] is unset and lets the daemon resolve the config/model default.
Switching models carried the old level over (coerced), while the TUI model picker pre-selects the target model's own default level.

What changed

All prompt paths now submit rawState.thinking verbatim; the coercion helpers (coerceThinkingForModel, coercePromptThinking) are removed.
rawState.thinking is ThinkingLevel | undefined. An absent/invalid stored value means "no preference": prompts omit the thinking field and the daemon resolves the default (the kap-server prompt route already treats thinking as optional). Only concrete levels are persisted to localStorage.
thinkingLevelForModelSwitch now mirrors the TUI picker: switching onto a different model pre-selects that model's default level; re-selecting the current model keeps the level untouched.
Composer, mobile settings sheet, and /status display the effective level via a new effectiveThinkingLevel helper (stored value, else the model default), and /thinking cycles from it.
Tests that locked in the old coercion were replaced with verbatim/undefined-semantics cases. Includes a patch changeset for @moonshot-ai/kimi-code (web is bundled into the CLI artifact).

Checklist

I have read the CONTRIBUTING document.
I have linked a related issue, or explained the problem above.
I have added tests that prove my feature works.
Ran gen-changesets skill, or this PR needs no changeset.
Ran gen-docs skill, or this PR needs no doc update (behavior fix only; no documented behavior changes).

changeset-bot · 2026-07-14T06:02:04Z

🦋 Changeset detected

Latest commit: 94dd9bf

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@moonshot-ai/kimi-code	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

pkg-pr-new · 2026-07-14T06:04:03Z

pnpm dlx https://pkg.pr.new/@moonshot-ai/kimi-code@9dec241

npx https://pkg.pr.new/@moonshot-ai/kimi-code@9dec241

commit: 9dec241

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7fad48053b

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-07-14T06:06:57Z

+  model: ModelThinkingInfo | undefined,
+  level: ThinkingLevel | undefined,
+): ThinkingLevel {
+  return level ?? defaultThinkingLevelFor(model);


Use daemon-resolved thinking for unset preference

When level is undefined, this assumes the daemon will use the model default, but the daemon resolver honors [thinking] config first (packages/agent-core-v2/src/app/model/thinking.ts:79-93), so enabled=false or a configured effort can override the model default. In a fresh web client with no persisted thinking, prompts now omit thinking, meaning Composer/mobile /status can display and /thinking can cycle from a level that is not what will actually run; please use the daemon-resolved status/config value for the unset case instead of deriving it solely from the model.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-07-14T06:06:57Z

      if (nextThinking !== prevThinking) {
-        rawState.thinking = prevThinking;
-        saveThinkingToStorage(prevThinking);
+        applyThinkingLevel(prevThinking);


Clear persisted thinking when rolling back to unset

When a model switch fails after prevThinking was unset, this rollback calls applyThinkingLevel(undefined), but applyThinkingLevel only writes localStorage for concrete levels; the target model default saved by the optimistic call remains in storage. Reproduces by switching models from a fresh/no-preference state while the profile POST fails: the UI rolls back for this session, but after reload prompts send the stale explicit effort instead of omitting thinking.

Useful? React with 👍 / 👎.

Align kimi-web's thinking-level handling with the TUI: - Submit the stored level as-is on every prompt path (prompt, steer, skill activation, BTW side chat) instead of coercing it onto the target model's declared efforts. - No stored preference (undefined) instead of a hardcoded 'high' default: prompts omit the thinking override and the daemon resolves the config/model default, same as an unset [thinking] in the TUI. - Model switcher pre-selects the target model's own default level when switching models; re-selecting the current model keeps the level. - Display the effective level (stored value, else the model default) in the composer, mobile sheet, and /status panel.

The stub daemon (dev/stub-daemon.mjs) no longer exists, so the dev:stub npm script and its docs references were dead weight.

…ally - With no stored preference, loadModels() pins the active model's catalog default_effort as a concrete in-memory value, so what the UI shows, what prompts submit, and what the session runs always agree. localStorage stays reserved for levels the user picked. - setThinking and model switches now also write the daemon-wide [thinking] config (same mapping as the TUI's thinkingEffortToConfig), so sessions created by other clients inherit the pick.

wbxl2000 · 2026-07-14T07:03:02Z

@codex

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9dec241470

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-07-14T07:06:30Z

+      if (rawState.thinking === undefined) {
+        const active = modelById(currentModelId());
+        if (active !== undefined) rawState.thinking = defaultThinkingLevelFor(active);


Defer default pin until the active session is known

When there is no stored thinking preference, this pins rawState.thinking during loadModels(), but the first-load flow calls modelProvider.loadModels() before sessions are loaded and before a deep-linked or auto-selected session is chosen. In that path currentModelId() only sees the global default model, so opening an existing session whose model differs from the default can store e.g. off and later submitPrompt sends that explicit value instead of letting the daemon resolve the session/model default.

Useful? React with 👍 / 👎.

chatgpt-codex-connector Bot reviewed Jul 14, 2026

View reviewed changes

wbxl2000 force-pushed the fix/web-thinking-level-verbatim branch from 7fad480 to 8fe65ef Compare July 14, 2026 06:24

wbxl2000 changed the base branch from main to fix/v2-model-catalog-efforts July 14, 2026 06:24

wbxl2000 added 3 commits July 14, 2026 15:00

chore(web): remove the dead dev:stub script

1513d29

The stub daemon (dev/stub-daemon.mjs) no longer exists, so the dev:stub npm script and its docs references were dead weight.

wbxl2000 force-pushed the fix/web-thinking-level-verbatim branch from 94dd9bf to 9dec241 Compare July 14, 2026 07:01

wbxl2000 changed the base branch from fix/v2-model-catalog-efforts to main July 14, 2026 07:01

wbxl2000 merged commit 0f64b4d into main Jul 14, 2026
13 checks passed

wbxl2000 deleted the fix/web-thinking-level-verbatim branch July 14, 2026 07:04

chatgpt-codex-connector Bot reviewed Jul 14, 2026

View reviewed changes

github-actions Bot mentioned this pull request Jul 14, 2026

ci: release packages #1682

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(web): submit thinking level verbatim and drop the hardcoded default - #1673

fix(web): submit thinking level verbatim and drop the hardcoded default#1673
wbxl2000 merged 3 commits into
mainfrom
fix/web-thinking-level-verbatim

wbxl2000 commented Jul 14, 2026

Uh oh!

changeset-bot Bot commented Jul 14, 2026 •

edited

Loading

Uh oh!

pkg-pr-new Bot commented Jul 14, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Jul 14, 2026

Uh oh!

chatgpt-codex-connector Bot Jul 14, 2026

Uh oh!

wbxl2000 commented Jul 14, 2026

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Jul 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

wbxl2000 commented Jul 14, 2026

Related Issue

Problem

What changed

Checklist

Uh oh!

changeset-bot Bot commented Jul 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🦋 Changeset detected

Uh oh!

pkg-pr-new Bot commented Jul 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jul 14, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Jul 14, 2026

Choose a reason for hiding this comment

Uh oh!

wbxl2000 commented Jul 14, 2026

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jul 14, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

changeset-bot Bot commented Jul 14, 2026 •

edited

Loading

pkg-pr-new Bot commented Jul 14, 2026 •

edited

Loading