refactor(agent): remove legacy command tools

docs/architecture/agent-sdk-product-architecture.md

@@ -341,7 +341,7 @@ await using client = await AgentClient.start({ cwd });
 // 常见的一次 Query；返回可迭代、可取消、可关闭的句柄。
 await using query = await client.query({
   prompt: "Find and fix the failing test",
-  allowedTools: ["Read", "Edit", "Bash"],
+  allowedTools: ["Read", "Edit", "ExecCommand"],
 });
 for await (const message of query) {
   // Message / Event / Result

docs/development/ui-testids-CN.md

@@ -209,8 +209,8 @@
 | Chat shell 命令文本 | `chat-shell-command-text` | Shell 命令文本节点。 |
 | Chat shell 命令输出 | `chat-shell-command-output` | Shell 命令 stdout/stderr 或实时输出区域。 |
 | Chat shell 命令退出码 | `chat-shell-command-exit-code` | 退出码节点。包含 `data-exit-code` 和 `data-status`。 |
-| Chat shell 工具卡片 | `chat-shell-tool-card` | Bash 的外层 FlowToolCard wrapper。包含 `data-tool-name` 和 `data-tool-card-id`。 |
-| Chat shell 工具打开面板按钮 | `chat-shell-tool-open-panel` | 存在 terminal session 时，从 Bash ToolCard 打开关联终端面板。 |
+| Chat shell 工具卡片 | `chat-shell-tool-card` | ExecCommand 的外层 FlowToolCard wrapper。包含 `data-tool-name` 和 `data-tool-card-id`。 |
+| Chat shell 工具打开面板按钮 | `chat-shell-tool-open-panel` | 存在 terminal session 时，从 ExecCommand 卡片打开关联终端面板。 |
 | Chat browser 工具卡片 | `chat-browser-tool-card` | WebFetch 的外层 FlowToolCard wrapper。包含 `data-tool-name` 和 `data-tool-card-id`。 |
 | Chat 文件变更卡片 | `chat-file-change-card` | 文件操作卡片根节点。包含 `data-status`、`data-action`、`data-path` 和 `data-expanded`。 |
 | Chat 文件变更展开按钮 | `chat-file-change-toggle` | 文件操作卡片的展开/收起点击目标。 |
@@ -299,7 +299,7 @@
 
 说明：
 
-- 独立 xterm 终端没有结构化的逐命令历史 DOM。测试应使用 `shell-command-output` 断言终端渲染输出，使用 `chat-shell-command-*` 断言结构化 Bash ToolCard。
+- 独立 xterm 终端没有结构化的逐命令历史 DOM。测试应使用 `shell-command-output` 断言终端渲染输出，使用 `chat-shell-command-*` 断言结构化 ExecCommand 卡片。
 - `shell-command-copy` 当前未暴露，因为活动终端复制能力基于选择/右键上下文菜单，并不是稳定可见按钮。
 
 ## Browser Panel

docs/development/ui-testids.md

@@ -208,7 +208,7 @@ Avoid adding IDs to these surfaces unless there is a clear automated workflow.
 | Chat shell command text | `chat-shell-command-text` | Shell command text node. |
 | Chat shell command output | `chat-shell-command-output` | Shell command stdout/stderr or live output area. |
 | Chat shell command exit code | `chat-shell-command-exit-code` | Exit code node. Includes `data-exit-code` and `data-status`. |
-| Chat shell tool card | `chat-shell-tool-card` | Outer FlowToolCard wrapper for Bash. Includes `data-tool-name` and `data-tool-card-id`. |
+| Chat shell tool card | `chat-shell-tool-card` | Outer FlowToolCard wrapper for ExecCommand. Includes `data-tool-name` and `data-tool-card-id`. |
 | Chat shell tool open panel | `chat-shell-tool-open-panel` | Opens the associated terminal panel when a terminal session is available. |
 | Chat browser tool card | `chat-browser-tool-card` | Outer FlowToolCard wrapper for WebFetch. Includes `data-tool-name` and `data-tool-card-id`. |
 | Chat file change card | `chat-file-change-card` | File operation card root. Includes `data-status`, `data-action`, `data-path`, and `data-expanded`. |
@@ -299,7 +299,7 @@ Avoid adding IDs to these surfaces unless there is a clear automated workflow.
 
 Notes:
 
-- The standalone xterm terminal does not expose a structured per-command history DOM. Tests should use `shell-command-output` for rendered terminal output and `chat-shell-command-*` for structured Bash ToolCard assertions.
+- The standalone xterm terminal does not expose a structured per-command history DOM. Tests should use `shell-command-output` for rendered terminal output and `chat-shell-command-*` for structured ExecCommand card assertions.
 - `shell-command-copy` is not currently exposed because the active terminal copy action is context-menu/selection driven rather than a stable visible button.
 
 ## Browser Panel

docs/features/session-runtime-usage-report-design.md

@@ -107,8 +107,7 @@ Models
 gpt-5.4:          8 req, 183,420 input, 21,908 output
 
 Tools
-Bash:             14 calls, 2m 31s, 2 errors
-Git:              5 calls, 42s
+ExecCommand:      19 calls, 3m 13s, 2 errors
 Write/Edit:       7 calls, 1m 08s
 ```
 
@@ -151,9 +150,9 @@ Current implemented Markdown shape:
 | gpt-5.4 | 8 | 183,420 | 21,908 | 205,328 |
 
 ### Slowest Work
-1. Bash `pnpm run build:web` - 1m 42s
+1. ExecCommand `pnpm run build:web` - 1m 42s
 2. Context compression - 28s
-3. Git `fetch origin main` - 19s
+3. ExecCommand `git fetch origin main` - 19s
 ```
 
 The detailed visual report exists alongside the Markdown snapshot. It uses the structured DTO when present and falls back to the Markdown snapshot for historical/local-only reports.
@@ -316,7 +315,7 @@ Required change:
 
 Missing:
 
-- Git can happen through the dedicated Git tool or through Bash/terminal commands.
+- Git now runs through ExecCommand; persisted legacy sessions may still contain dedicated Git or Bash calls.
 
 Required change:
 
@@ -1384,8 +1383,8 @@ Steps:
 1. Persist or emit queue wait, preflight, confirmation wait, execution, and total duration for completed tools.
 2. Include best-effort total duration for failed and cancelled tools.
 3. Add a report-only classifier for tool categories.
-4. Classify dedicated Git tool calls as `git`.
-5. Classify terminal calls as `git` only when the normalized command clearly invokes Git.
+4. Classify legacy dedicated Git tool calls as `git`.
+5. Classify ExecCommand and legacy terminal calls as `git` only when the normalized command clearly invokes Git.
 6. Classify file operations by tool name and snapshot operation metadata.
 7. When SubagentScheduler or budget governance events exist, consume their queued/running/retry/backoff summaries as runtime facts instead of inferring them from tool names.
 
@@ -1407,8 +1406,8 @@ Risks and mitigations:
 
 Verification:
 
-- Unit tests for Git tool classification.
-- Unit tests for Bash Git command classification and false positives.
+- Unit tests for legacy Git tool classification.
+- Unit tests for ExecCommand Git command classification and false positives.
 - Existing tool lifecycle tests still pass.
 
 ### Task 10: File-change report integration