feat(acp-adapter): expose thinking effort levels in config options

[cnctem]

Related Issue

#1947

Problem

The ACP thinking config option currently exposes only a binary off/on toggle, even for models that support multiple thinking effort levels (low, medium, high, etc.). The VS Code extension's ThinkingButton component already supports per-model effort selection, but the ACP adapter does not, leaving editor integrations like Zed with a coarser control than the first-party UI.

What changed

This PR aligns the ACP adapter's thinking control with apps/vscode/webview-ui/src/components/ThinkingButton.tsx:

• Effort dropdown: when a model advertises supportEfforts, the thinking option now renders ['off', ...efforts] (e.g. off / low / medium / high), matching ThinkingButton's effort mode.
• Binary fallback: models without effort lists keep the existing off / on select, equivalent to ThinkingButton's switch mode.
• Locked-on state: alwaysThinking models collapse to a single on entry, equivalent to ThinkingButton's always-on mode.
• Capitalized labels: option labels use the same capitalization style as ThinkingButton (Off, On, Low, Medium, …).

Implementation details:

• packages/acp-adapter/src/model-catalog.ts: expose supportEfforts on AcpModelEntry, sourced from the model alias config or Anthropic model profile inference.
• packages/acp-adapter/src/config-options.ts: buildThinkingOption now renders an effort dropdown when supportEfforts is present, falling back to binary off/on otherwise.
• packages/acp-adapter/src/session.ts: adapter-side thinking state is now tracked as an effort string instead of a boolean, and setThinking accepts effort strings with 'on' normalized to the model default.
• packages/acp-adapter/src/server.ts: resolveCurrentThinkingEffort returns the actual effort string; setSessionConfigOption({ configId: 'thinking', value }) passes the effort through.

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.

[changeset-bot]

🦋 Changeset detected

Latest commit: b60a97d

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

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4d31ea1588

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]

Show effort choices for always-on effort models

When a catalog row has both alwaysThinking and supportEfforts, this early return runs before the effort-dropdown branch, so ACP exposes only a single on choice. Such rows are real: effectiveModelAlias assigns always_thinking for profiles with canDisableThinking === false while still copying profile.efforts (packages/agent-core/src/config/model.ts:41-56), and managed/open-platform models can also be supports_thinking_type='only' with efforts. In those environments the new ACP surface still hides the low/medium/high/... selector for effort-capable models; the always-on case should omit only off, not collapse all efforts to on.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Don't render carried thinking effort as off

When currentEffort is any non-off value that is not declared by the currently selected model, this reports currentValue: 'off' without actually disabling thinking. For example, after selecting max on one model and switching to a Kimi model whose supportEfforts are only ['low','high'], AcpSession.setModel preserves the non-off adapter state while agent-core normalizes unsupported Kimi efforts to the target model default (packages/agent-core/src/agent/config/thinking.ts:79-81); the ACP snapshot then tells the client thinking is off even though the session will run with thinking on. Prefer the target/default effort for non-off carried values, or refresh the adapter state from the session after the model switch.

Useful? React with 👍 / 👎.

[cnctem]

针对 Review 的两点修改如下：

1. always-thinking + supportEfforts 现在展示 effort 选项

文件：packages/acp-adapter/src/config-options.ts

// 修改前
if (alwaysThinking) {
  return { currentValue: 'on', options: [{ value: 'on', name: 'On' }] };
}

// 修改后
if (alwaysThinking) {
  if (efforts.length > 0) {
    // 有 effort 列表时展示全部 effort，只去掉 off
    return {
      currentValue: normalizedEffort,
      options: efforts.map((effort) => ({ value: effort, name: label(effort) })),
    };
  }
  // 没有 effort 列表时才展示单个 On
  return { currentValue: 'on', options: [{ value: 'on', name: 'On' }] };
}

匹配 ThinkingButton 的 alwaysOn 行为：options = alwaysOn ? efforts : ['off', ...efforts]。

2. 模型切换后遗留的 effort 映射到默认 effort，不再显示为 off

文件：packages/acp-adapter/src/config-options.ts 的 buildSessionConfigOptions

const efforts = currentModelEntry?.supportEfforts;
let normalizedEffort = currentThinkingEffort;
if (
  normalizedEffort === 'off' ||
  (efforts !== undefined &&
    efforts.length > 0 &&
    !efforts.includes(normalizedEffort))
) {
  normalizedEffort =
    normalizedEffort === 'off'
      ? 'off'
      : (currentModelEntry?.defaultThinkingEffort ?? 'on');
}
if (alwaysThinking && normalizedEffort === 'off') {
  normalizedEffort = currentModelEntry?.defaultThinkingEffort ?? 'on';
}

逻辑：

• off 保持 off
• 'on' 或任何不在当前模型 supportEfforts 里的非 off 值 → 当前模型的 defaultThinkingEffort
• always-thinking 模型永远不会展示 off

对应新增测试

文件：packages/acp-adapter/test/config-options.test.ts

• shows effort choices without an off entry for always-thinking models with supportEfforts
• falls back to the first effort for always-thinking models when the current effort is unsupported
• maps a carried unsupported effort to the target model default effort

[cnctem]

#1992 已经实现该功能，故关闭