Release: develop を main にマージ

[otomatty]

概要

develop の最新を main に取り込むリリース用 PR です。Tauri / Claude Code 連携、AI チャット・エディタ拡張、MCP、HTML サニタイズの強化、各種 UI 改善などを含みます。

Release PR merging the current develop into main.

変更点

• デスクトップ / Claude Code: Tauri 2.0、サイドカー・デスクトップコマンド、エージェント・スラッシュコマンド、Claude Code マルチステップワークフローパネル、MCP サーバー連携
• AI: Claude Code プロバイダとプロバイダ統合、サイドパネル・エージェントチャット、AI 設定モード、ワークスペース連携ノート（@file: / cwd）
• エディタ / UI: 統合プロバイダ・実行可能コードブロック、WikiLink ホバーカードプレビュー、YouTube 埋め込み、モバイルでカード上スクロール不具合の修正、ページ一覧のコンテンツプレビュー同期
• セキュリティ: カスタム HTML サニタイザを DOMPurify に置き換え
• 依存関係: @anthropic-ai/sdk などの更新

変更の種類

• ✨ 新機能 (New feature)
• 🐛 バグ修正 (Bug fix)
• 🔧 ビルド/CI (Build/CI)
• 💥 破壊的変更 (Breaking change)
• 📝 ドキュメント (Documentation)
• 🎨 スタイル/リファクタリング (Style/Refactor)
• 🧪 テスト (Tests)

テスト方法

1. CI（lint / test 等）がグリーンであることを確認する
2. マージ後、本番相当環境で主要フロー（ログイン、ノート編集、AI チャット、デスクトップ利用時は Tauri 周り）をスポット確認する

チェックリスト

• テストがすべてパスする（CI を確認）
• Lint エラーがない（CI を確認）
• 必要に応じてドキュメントを更新した
• コミットメッセージが Conventional Commits に従っている

マージ方法

プロジェクト方針に合わせ、Create a merge commit でのマージを推奨します。

スクリーンショット（UI 変更がある場合）

UI 変更を含みます。必要に応じてリリースノート用に主要画面のキャプチャを添付してください。

関連 Issue

---

Summary by CodeRabbit

• New Features

  • デスクトップアプリ（Tauri）対応とClaude Codeサイドカー統合
  • 実行可能なコードブロック（実行・解釈機能）とワークフロー実行パネル
  • MCPサーバーの追加・管理、ノート単位ワークスペース連携、ファイルプレビュー
  • YouTube埋め込み・外部ファイル参照・スラッシュエージェントコマンド

• 改善

  • AIモード選択／モデル選択UI・ヘッダ表示の更新（モードバッジ、MCPステータス表示）
  • コピー／ノート挿入操作、モバイル長押しなど操作性向上
  • ページのコンテンツプレビュー生成と永続化の改善

• Bug Fixes

  • サーバー書き込みをトランザクション化し整合性を向上

[coderabbitai]

📝 Walkthrough

Walkthrough

Tauri デスクトップアプリと Claude Code サイドカーの導入、TipTap エディタ拡張（実行可能コードブロック、MCP リソース、スラッシュエージェント等）、ワークフロー実行・管理、ワークスペース連携、多プロバイダ AI 統合、および関連 UI/テスト/ローカライズの大規模追加。

Changes

Cohort / File(s)	Summary
Tauri & sidecar 基盤 src-tauri/Cargo.toml, src-tauri/build.rs, src-tauri/tauri.conf.json, src-tauri/src/lib.rs, src-tauri/src/main.rs, src-tauri/capabilities/default.json, src-tauri/icons/...	Tauri アプリ設定、プラグイン・権限・アイコン、外部バイナリ（claude-sidecar）を含むデスクトップランタイムを追加。
claude-sidecar パッケージ & プロトコル packages/claude-sidecar/src/*, packages/claude-sidecar/package.json, packages/claude-sidecar/tsconfig.json, packages/claude-sidecar/vitest.config.ts, packages/claude-sidecar/src/index.ts, scripts/*.mjs	JSONL プロトコル、クエリ実行/中止/ステータス/モデル列挙、ビルド/ensure スクリプト、テスト設定を追加。
Tauri ⇄ sidecar ブリッジ（フロント） src/lib/claudeCode/*, src/lib/claudeCode/bridge.ts, src/lib/claudeCode/types.ts, src/lib/claudeCode/streamClaudeQuery.ts, src/lib/claudeCode/runQueryToCompletion.ts	フロントでの Tauri ブリッジ、イベント購読、ストリーミング集約ロジックを導入。
Tauri ⇄ sidecar ブリッジ（Rust） src-tauri/src/claude_sidecar.rs	Rust 側で sidecar プロセス管理、JSONL RPC ルーティング、Tauri コマンド公開を追加。
AI サービス層 & プロバイダ統合 src/lib/aiService.ts, src/lib/aiServiceServer.ts, src/lib/aiServiceDirectProviders.ts, src/lib/aiServiceModels.ts, src/lib/aiClient.ts	AI 呼び出しの統一レイヤーをリファクタ／拡張し、claude-code を経路に追加、モデル取得と使用量 API を分離。
プロバイダ実装 src/lib/aiProviders/*.ts, src/lib/aiProviders/registry.ts, src/lib/aiProviders/types.ts	OpenAI/Anthropic/Google/provide registry と claude-code provider の追加（統一インターフェース）。
エディタ拡張（TipTap）全般 src/components/editor/..., src/components/editor/extensions/*, src/components/editor/TiptapEditor/*	ExecutableCodeBlock, FileReference, McpResource, YouTubeEmbed, StorageImage 等の拡張、スラッシュメニュー再構成、ホバーカード、パス補完、insertAtCursorRef 等を追加・連携。
エージェントスラッシュコマンド src/lib/agentSlashCommands/*	エージェントコマンドパーサ、プロンプトビルダ、実行フロー、挿入ユーティリティ、フックを追加（/analyze 等）。
実行可能コード実行基盤 src/lib/executableCode/*, src/components/editor/executableCodeBlock/*	実行プロンプトの構築、Claude 経由での実行・解釈、NodeView UI、ツールバー、確認ダイアログ、出力パースを追加。
ワークフロー機能 src/lib/workflow/*, src/hooks/useWorkflowDraft.ts, src/hooks/useWorkflowRunSession.ts, src/hooks/useWorkflowPanelLogic.ts, src/components/ai-chat/*Workflow*	ワークフロードラフト管理、実行セッション（中断/再開）、プロンプト生成、進捗フォーマット、ワークフローパネル UI を導入。
MCP サーバー統合 & 管理 UI src/components/settings/McpServerSettings.tsx, src/components/settings/*McpServerForm*.tsx, src/lib/mcpServerImportHelpers.ts, src/components/settings/mcpServerFormInitialState.ts	MCP サーバー CRUD、インポート/検証、フォーム状態と UI を追加。
ワークスペース連携 / ファイルプレビュー src/lib/noteWorkspace/*, src/contexts/NoteWorkspaceContext.tsx, src/components/note/*	note に紐づくワークスペース登録・永続化、Tauri 経由ファイル読み取り・一覧、FilePreview イベント/ダイアログ、管理ツールバーを実装。
AI チャット UI 変更 src/components/ai-chat/*, src/contexts/AIChatContext.tsx	モードバッジ、モデルセレクタ拡張、MCP ステータス表示、メッセージ挿入/コピーアクション、workflow タブを追加。
設定周り・i18n・ドキュメント src/components/settings/*, src/i18n/locales/*, .claude/settings.local.json, .mcp.json, .env.example, AGENTS.md, .cursor/skills/handle-pr-review/SKILL.md, .claude/commands/review-pr-comments.md (削除)	AI 設定の 3 モード化、Claude Code 前提チェック、i18n 追加、PR レビュー運用のドキュメント更新。
ビルド/パッケージ等 package.json, knip.json, .gitignore, scripts/build-claude-sidecar.mjs, scripts/ensure-claude-sidecar-bin.mjs	npm スクリプト追加（sidecar/tauri）、依存追加、ワークスペース・サイドカーのビルド/配置スクリプト、ignore ルール更新。
サニタイズ & WebClipper src/lib/webClipper/sanitizeHtml.ts, src/lib/webClipper/sanitizeHtmlConfig.ts, src/components/editor/utils/urlTransform.ts	DOMPurify ベースのサニタイズへ移行、Gyazo/YouTube URL 変換ユーティリティ追加。
テスト追加・修正 src/**/*.test.ts?(x)（多数）	プロバイダ、sidecar ブリッジ、エディタ機能、ワークフロー、parse/ユーティリティ等のユニット/統合テスト多数を追加。

Sequence Diagram(s)

sequenceDiagram
    autonumber
    participant UI as App (Renderer)
    participant Tauri as Tauri (Rust)
    participant Sidecar as Sidecar (JSONL proc)
    participant Claude as Claude SDK

    UI->>Tauri: invoke `claude_query` (prompt, options)
    Tauri->>Sidecar: write JSONL request to stdin
    Sidecar->>Claude: run query via Claude SDK / agent
    Claude-->>Sidecar: stream chunk events (tool start/stop, text deltas)
    Sidecar-->>Tauri: emit JSONL response lines (stream-chunk / tool-use / mcp-status)
    Tauri-->>UI: emit Tauri events (claude-stream-chunk / claude-tool-use-*)
    UI->>Tauri: (optionally) invoke `claude_abort`
    Tauri->>Sidecar: write abort JSONL
    Sidecar->>Claude: abort underlying request
    Sidecar-->>Tauri: emit stream-complete / error
    Tauri-->>UI: emit completion/error event

Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~75 分

Possibly related PRs

• feat(tauri): add claude code sidecar bridge and desktop commands (#456) #465 — Tauri sidecar / Claude Code bridge と packages/claude-sidecar の導入（同一サイドカー実装とビルドスクリプトに強い関連）。
• feat(editor): agent slash commands via claude code sidecar (#460) #475 — エージェントスラッシュコマンドとスラッシュメニュー再構成（agent slash 周りのロジック／プロンプト生成に重複）。
• feat: mcp server integration for claude code on desktop #478 — MCP サーバー統合と McpResource ノード／設定 UI（MCP 設定・状態管理の重複）。

Suggested labels

enhancement

"🐰 こりゃ新機能いっぱい、コード跳ねるよ！
サイドカー走らせ、ワークフロー回し、
エディタに魔法を、MCP もつなげて、
デスクトップで Claude が歌うよ ✨"

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch develop

[gemini-code-assist]

Code Review

This pull request introduces a desktop-native workflow feature using Tauri and Claude Code, including a sidecar process for secure communication, a new workflow UI panel, and support for MCP resources. I have reviewed the changes and found an issue in the text extraction logic for content previews in server/hocuspocus/src/index.ts, where improper newline insertion can lead to fragmented text. A suggestion has been provided to use spaces instead of newlines for block elements to ensure better formatting.

[gemini-code-assist]

現在のテキスト抽出ロジックでは、インライン要素（<strong>など）の後にも改行(\n)が追加されてしまい、不自然なスペースがプレビューや検索用テキストに含まれる可能性があります。

例えば、<p>Hello <strong>world</strong>!</p> は Hello world\n! のように変換されてしまいます。

ブロック要素とインライン要素を区別するのは難しいですが、ここでは改行の代わりにスペースを追加することで、後続の replace(/\s+/g, " ") 処理で適切に整形されるようにするのが安全です。

Suggested change

	} else if (child instanceof Y.XmlElement) {
	text += extractTextFromFragment(child) + "\n";
	}
	} else if (child instanceof Y.XmlElement) {
	text += extractTextFromFragment(child) + " ";
	}

[otomatty]

ご指摘のとおりでした。インライン要素のあとに一律 \n を付けないよう、bold / italic などインライン相当の nodeName を列挙し、その XmlElement の後はスペース、ブロック要素の後は改行とする実装に変更しました（server/hocuspocus/src/extractPlainTextFromYXml.ts）。単体テストも追加済みです。対応は develop 向け PR #493 経由でマージ済みです。

Implemented inline vs block suffixes in extractPlainTextFromYXml.ts (not a blanket space for every element), with tests. Landed via #493.

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

[otomatty]

PR #492 のインラインレビュー（Gemini / CodeRabbit）に対する修正を develop 向け PR #493 で出しています。マージ後に本リリース PR の差分を更新してください。

• fix: address PR #492 review comments (Y.Xml extract, testConnection, useRef) #493

内容: Hocuspocus の Y.Xml プレーンテキスト抽出（インライン要素で不要改行を避ける）、testConnection の claude-code + 空キー、AIChatWikiLink の useRef 初期値。

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (3)

server/hocuspocus/src/extractPlainTextFromYXml.test.ts (1)

54-62: 境界値テストの追加を検討してください（任意）

buildContentPreview のテストに、ちょうど 120 文字の境界値ケースを追加すると、切り詰めロジックの正確性をより確実に検証できます。

🧪 追加テストの提案

 describe("buildContentPreview", () => {
   it("collapses whitespace and truncates", () => {
     expect(buildContentPreview("  a \n b  ")).toBe("a b");
     const long = "x".repeat(200);
     const prev = buildContentPreview(long);
     expect(prev.endsWith("...")).toBe(true);
     expect(prev.length).toBeLessThanOrEqual(124);
   });
+
+  it("does not truncate text exactly at max length", () => {
+    const exact = "x".repeat(120);
+    expect(buildContentPreview(exact)).toBe(exact);
+    expect(buildContentPreview(exact).endsWith("...")).toBe(false);
+  });
+
+  it("returns empty string for empty input", () => {
+    expect(buildContentPreview("")).toBe("");
+    expect(buildContentPreview("   ")).toBe("");
+  });
 });

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@server/hocuspocus/src/extractPlainTextFromYXml.test.ts` around lines 54 - 62,
Add a boundary test for buildContentPreview that supplies an input string
exactly at the truncation threshold (e.g., 120 characters) to verify the cut-off
logic; create a 120-char string (using repeat or concatenation), call
buildContentPreview with it, and assert that the result is not suffixed with
"..." and that its length meets the expected non-truncated length (or matches
the original preview length rules), so you validate the edge-case behavior of
buildContentPreview.

server/hocuspocus/src/extractPlainTextFromYXml.ts (1)

10-21: インライン要素のカスタマイズについて（任意の改善）

現在の INLINE_XML_ELEMENT_NAMES は一般的な TipTap マークをカバーしていますが、カスタムマークやエディタ拡張が追加された場合、不自然な改行が発生する可能性があります。将来的には、この Set を設定可能にするか、フォールバック検出ロジック（例：子要素が XmlText のみの場合はインラインとみなす）を検討してください。

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@server/hocuspocus/src/extractPlainTextFromYXml.ts` around lines 10 - 21, The
hardcoded INLINE_XML_ELEMENT_NAMES Set can cause incorrect block/inline
detection for custom marks; update the code to allow configuration and add a
fallback that treats an XmlXMLElement as inline when its children are all
XmlText (or empty) — expose INLINE_XML_ELEMENT_NAMES as a configurable parameter
(or accept an optional set in extractPlainTextFromYXml and/or its caller), and
inside the detection logic in extractPlainTextFromYXml use: first check the
configured set, and if not present, treat elements as inline when every child
node is of type XmlText (or there are no child elements); ensure to reference
and update the INLINE_XML_ELEMENT_NAMES constant and the element-kind check in
extractPlainTextFromYXml accordingly.

server/hocuspocus/src/index.ts (1)

199-200: 空文字列と null の扱いの一貫性について（任意の改善）

clipAndCreate.ts では contentText || null パターンを使用していますが、ここでは空文字列をそのまま保存しています。動作上は問題ありませんが、他のコードとの一貫性を保つ場合は以下の変更を検討してください。

♻️ 一貫性のための修正案

-  const contentText = extractTextFromYXml(document.getXmlFragment("default"));
-  const contentPreview = buildContentPreview(contentText);
+  const rawText = extractTextFromYXml(document.getXmlFragment("default"));
+  const contentText = rawText.trim() || null;
+  const contentPreview = contentText ? buildContentPreview(rawText) : null;

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@server/hocuspocus/src/index.ts` around lines 199 - 200, The code currently
assigns contentText = extractTextFromYXml(document.getXmlFragment("default"))
and leaves empty string as-is whereas other code (clipAndCreate.ts) uses
contentText || null for consistency; update this location to normalize empty
content to null (e.g., set contentText = extractTextFromYXml(...) || null) or
adjust buildContentPreview to accept nulls, and ensure contentPreview is derived
from that normalized value so the module uses the same empty-string/null
convention as clipAndCreate.ts; reference extractTextFromYXml,
buildContentPreview, contentText, and contentPreview when making the change.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/lib/aiClient.ts`:
- Around line 8-9: The exported type AIClient and the exported
ConnectionTestResult need bilingual JSDoc/TSDoc comments (Japanese and English)
per repo guidelines; update the JSDoc above the AIClient type declaration and
above the ConnectionTestResult declaration to include both a short Japanese
description and an English description, mention that AIClient is a union of SDK
clients (OpenAI, Anthropic, GoogleGenAI) and that ConnectionTestResult
represents the result shape for connection tests so reviewers can find them by
symbol names AIClient and ConnectionTestResult.

---

Nitpick comments:
In `@server/hocuspocus/src/extractPlainTextFromYXml.test.ts`:
- Around line 54-62: Add a boundary test for buildContentPreview that supplies
an input string exactly at the truncation threshold (e.g., 120 characters) to
verify the cut-off logic; create a 120-char string (using repeat or
concatenation), call buildContentPreview with it, and assert that the result is
not suffixed with "..." and that its length meets the expected non-truncated
length (or matches the original preview length rules), so you validate the
edge-case behavior of buildContentPreview.

In `@server/hocuspocus/src/extractPlainTextFromYXml.ts`:
- Around line 10-21: The hardcoded INLINE_XML_ELEMENT_NAMES Set can cause
incorrect block/inline detection for custom marks; update the code to allow
configuration and add a fallback that treats an XmlXMLElement as inline when its
children are all XmlText (or empty) — expose INLINE_XML_ELEMENT_NAMES as a
configurable parameter (or accept an optional set in extractPlainTextFromYXml
and/or its caller), and inside the detection logic in extractPlainTextFromYXml
use: first check the configured set, and if not present, treat elements as
inline when every child node is of type XmlText (or there are no child
elements); ensure to reference and update the INLINE_XML_ELEMENT_NAMES constant
and the element-kind check in extractPlainTextFromYXml accordingly.

In `@server/hocuspocus/src/index.ts`:
- Around line 199-200: The code currently assigns contentText =
extractTextFromYXml(document.getXmlFragment("default")) and leaves empty string
as-is whereas other code (clipAndCreate.ts) uses contentText || null for
consistency; update this location to normalize empty content to null (e.g., set
contentText = extractTextFromYXml(...) || null) or adjust buildContentPreview to
accept nulls, and ensure contentPreview is derived from that normalized value so
the module uses the same empty-string/null convention as clipAndCreate.ts;
reference extractTextFromYXml, buildContentPreview, contentText, and
contentPreview when making the change.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: b01516f7-c303-49aa-9205-c243b5f1bb0e

📥 Commits

Reviewing files that changed from the base of the PR and between 324b91a and e0d93b1.

📒 Files selected for processing (6)

• server/hocuspocus/src/extractPlainTextFromYXml.test.ts
• server/hocuspocus/src/extractPlainTextFromYXml.ts
• server/hocuspocus/src/index.ts
• src/components/ai-chat/AIChatWikiLink.tsx
• src/lib/aiClient.test.ts
• src/lib/aiClient.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• src/components/ai-chat/AIChatWikiLink.tsx

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

公開型のJSDocを日英併記にしてください。

AIClient と ConnectionTestResult の新規JSDocが英語のみで、リポジトリのコメント規約を満たしていません。

✍️ 提案差分

-/** Union of SDK client types used for user-API-key mode. */
+/**
+ * ユーザーAPIキー方式で使用するSDKクライアントのユニオン型。
+ * Union of SDK client types used for user-API-key mode.
+ */
 export type AIClient = OpenAI | Anthropic | GoogleGenAI;
@@
-/** Result of a user-API-key connection test from the settings UI. */
+/**
+ * 設定画面におけるユーザーAPIキー接続テストの結果。
+ * Result of a user-API-key connection test from the settings UI.
+ */
 export interface ConnectionTestResult {

As per coding guidelines Export functions, types, and interfaces must include TSDoc or JSDoc comments. Comments and documentation must include both Japanese and English explanations where applicable.

Also applies to: 43-44

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/lib/aiClient.ts` around lines 8 - 9, The exported type AIClient and the
exported ConnectionTestResult need bilingual JSDoc/TSDoc comments (Japanese and
English) per repo guidelines; update the JSDoc above the AIClient type
declaration and above the ConnectionTestResult declaration to include both a
short Japanese description and an English description, mention that AIClient is
a union of SDK clients (OpenAI, Anthropic, GoogleGenAI) and that
ConnectionTestResult represents the result shape for connection tests so
reviewers can find them by symbol names AIClient and ConnectionTestResult.

[devin-ai-integration]

Devin Review found 4 new potential issues.

View 10 additional findings in Devin Review.

[devin-ai-integration]

🟡 McpResourceNodeView Retry button lacks disabled guard, allowing concurrent fetches that leak Tauri event listeners

The "Retry" button inside the error block (src/components/editor/McpResourceNodeView.tsx:213-220) does not set disabled={isLoading}, unlike the main Fetch button at line 177. If the user clicks Retry while a previous fetch is already in-flight (possible with quick double-clicks), a new set of onClaudeStreamComplete / onClaudeError listeners is registered and activeUnlistenRef.current is overwritten at line 142, orphaning the previous listeners. The old listeners are never unsubscribed, causing a resource leak. Each leaked listener also risks processing stale events for a different request id.

Suggested change

	<Button
	variant="link"
	size="sm"
	className="ml-2 h-auto p-0 text-xs"
	onClick={handleFetch}
	>
	{t("common.retry", "Retry")}
	</Button>
	<Button
	variant="link"
	size="sm"
	className="ml-2 h-auto p-0 text-xs"
	onClick={handleFetch}
	disabled={isLoading}
	>

---

Was this helpful? React with 👍 or 👎 to provide feedback.

[devin-ai-integration]

🟡 Sidecar assistant-message delta uses positional slice instead of prefix-aware diff, producing corrupt text when SDK message differs from streamed aggregation

In packages/claude-sidecar/src/handlers/query.ts:255, the delta for an assistant message is computed as text.slice(aggregated.length). This assumes the assistant message's full text always starts with exactly the same characters as aggregated (built from prior stream_event text deltas). If the SDK sends an assistant message whose text is shorter than aggregated (e.g. the model revised its output), the slice produces an empty string and the revision is silently dropped. If the text diverges from aggregated (same length, different prefix), the slice produces incorrect trailing characters that are emitted as a "delta" to the host, corrupting the displayed output.

Prompt for agents

In runQuery (packages/claude-sidecar/src/handlers/query.ts), the assistant-message handler at line 255 computes the delta as text.slice(aggregated.length). This is fragile because it assumes the full assistant text always starts with the exact same bytes as the aggregated stream content. A safer approach would be to check whether text starts with aggregated (e.g. text.startsWith(aggregated)) and only emit the suffix as a delta in that case. When the text does NOT start with aggregated, the handler should either replace the entire aggregated content and emit the full new text as a delta, or diff the two strings. The key functions involved are extractAssistantText and the msg.type === 'assistant' branch in runQuery.

---

Was this helpful? React with 👍 or 👎 to provide feedback.

[devin-ai-integration]

🔴 claudeCodeProvider closure shares mutable aborted/currentRequestId state across concurrent query() calls

In src/lib/aiProviders/claudeCodeProvider.ts:43-44, currentRequestId and aborted are declared as closure variables outside the query() generator. If two consumers call query() concurrently (e.g. chat + workflow panel), the second call resets aborted = false (line 63), overwrites currentRequestId (line 143), and steals all subsequent sidecar events. The first query silently stops receiving chunks and never completes. Calling abort() only cancels whichever query last wrote to currentRequestId.

Affected variables

• currentRequestId (line 43): overwritten by each claudeQuery() call
• aborted (line 44): reset at the start of each query()
• resolveWait (line 45): overwritten by the wait loop

Prompt for agents

The createClaudeCodeProvider factory in src/lib/aiProviders/claudeCodeProvider.ts uses closure-scoped variables (currentRequestId, aborted, resolveWait) that are shared across all invocations of query(). Move these variables inside the query() async generator so each call has its own isolated state. The abort() method needs a way to signal the active query - consider storing per-request AbortControllers in a Map keyed by request id, or having query() accept an AbortSignal (which it already does via the signal parameter) instead of relying on the shared aborted flag. The currentRequestId used by event listeners should also be scoped per invocation (e.g. captured by the listener closures registered in that specific call).

---

Was this helpful? React with 👍 or 👎 to provide feedback.

[devin-ai-integration]

🟡 Tauri event listener leak when early onClaude* registration throws in claudeCodeProvider

In src/lib/aiProviders/claudeCodeProvider.ts:80-118, all six Tauri event listeners (onClaudeStreamChunk, onClaudeStreamComplete, etc.) are registered sequentially before the try block at line 139. If, for example, onClaudeToolUseComplete (line 110) throws, the previously registered listeners (unlistenChunk, unlistenComplete, unlistenError, unlistenToolStart) are never cleaned up because the finally block at line 180 is inside a separate try that is never entered. The leaked listeners continue receiving events from all future sidecar queries.

Prompt for agents

In createClaudeCodeProvider (src/lib/aiProviders/claudeCodeProvider.ts), the six listener registrations at lines 80-118 happen before the try/finally block at lines 139-188. If any registration throws partway through, the already-registered listeners are leaked. Move all listener registrations inside the try block, and collect them into an array (const cleanups: Array<() => void> = []). In the finally block, iterate and call each cleanup. The streamClaudeQuery function in src/lib/claudeCode/streamClaudeQuery.ts already uses this pattern correctly (lines 110-183, cleanups array) and can serve as a reference.

---

Was this helpful? React with 👍 or 👎 to provide feedback.