Changes on 2026-08-15

1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Input:** A batch of 4 documentation file diffs (`claude-code-on-the-web.md`, `cli-reference.md`, `model-config.md`, `sub-agents.md`). * **Output Format:** Markdown with specific sections: Overall Summary, Key Themes, Impact Level, Action Items. * **Language:** Chinese (Simplified). 2. **Analyze Each Page:** * **Page 1: `claude-code-on-the-web.md`** * *Change:* Changed "Task tool" to "Agent tool" in the description of how Subagents work. * *Context:* "Claude can spawn them with the Task tool..." -> "Claude can spawn them with the Agent tool...". * *Meaning:* This is a terminology update. The underlying mechanism is likely the same, but the tool name used internally or by developers (and reflected in docs) has changed from `Task` to `Agent`. * **Page 2: `cli-reference.md`** * *Change:* Added a sentence to the `--allowedTools` description. * *Context:* "If you name one of the [task-tracking tools](/docs/en/tools-reference#task-tool-availability) here, Claude Code also opts the session in..." * *Meaning:* Clarification on behavior. Using `--allowedTools` with specific tracking tools implicitly opts the session into that feature set. * **Page 3: `model-config.md`** * *Change 1:* Added a sentence about diagnostic lines for unrecognized models. "Claude Code can still write the [unrecognized-model diagnostic line]... at request time, on every provider." * *Change 2:* Added a paragraph under `modelOverrides`. "To stop the `[claude-code:unrecognized_model]`... for an ID such as a gateway alias, add an entry with that ID as its value." * *Meaning:* Refinement in how model IDs are validated and reported, specifically regarding gateways and aliases. It gives developers a workaround to suppress warnings for valid-but-unknown gateway model aliases. * **Page 4: `sub-agents.md`** * *Change 1:* Updated the cleanup timing for completed background subagents. "...for the same window as the footer hint above." (Previously implied a different cleanup timing or lack of specific link to the footer hint). * *Change 2:* Clarified how to resume a stopped subagent. "While [that subagent's row is still in the subagent panel]..." (Added context condition). * *Meaning:* Usability and behavioral clarification. Developers now understand exactly how long background tasks persist and under what conditions they can interact with stopped tasks. 3. **Synthesize Key Themes:** * *Terminology Update:* "Task tool" -> "Agent tool". * *Configuration & Validation:* Model overrides and handling unrecognized model IDs (gateways). * *Behavioral Clarification:* Subagent lifecycle (cleanup, resumption) and CLI flags (`--allowedTools`). 4. **Determine Impact Level:** * Most changes are documentation improvements (clarifications, "how-to" details) and minor terminology updates. * The `modelOverrides` tip is helpful but not breaking. * The terminology change "Task tool" to "Agent tool" is significant for strictly interpreting logs or protocols, but likely reflects an internal refactor that happened earlier or is transparent to the user. * *Verdict:* **Low**. These are refinements and clarifications, not API changes or breaking behavioral shifts. 5. **Draft Action Items (in Chinese):** * Update references to "Task tool" to "Agent tool" if strict documentation matching is needed. * Use `modelOverrides` to suppress false warnings for gateway model aliases. * Review the `--allowedTools` behavior if relying on specific task-tracking tools. 6. **Draft the Response (Iterative refinement for Chinese):** * *Summary:* 本次更新主要涉及术语修正、模型配置与子代理行为的细节澄清。 * *Themes:* * **术语修正**：Subagent 生成工具从 "Task tool" 更新为 "Agent tool"。 * **模型配置优化**：明确了如何处理非识别模型 ID（特别是网关别名）的诊断信息及抑制方法。 * **行为与生命周期**：细化了后台子代理的清理时间逻辑及已停止子代理的恢复条件。 * **CLI 行为澄清**：解释了 `--allowedTools` 对任务跟踪工具的隐式影响。 * *Impact:* **Low (低)**。主要是文档描述的精确化，未发现破坏性变更。 * *Action Items:* * 如果使用网关模型别名收到误报警告，请配置 `modelOverrides` 以抑制警告。 * 若在代码或文档中引用了 "Task tool"，建议更新为 "Agent tool" 以保持一致。 * 注意后台子代理的列表保留时间现在与底部提示的时间窗口一致。 7. **Final Review against constraints:** * Markdown format? Yes. * Chinese language? Yes. * Concise? Yes. * Focus on what matters to developers? Yes (Terminology, suppression of warnings, lifecycle behavior). 8. **Construct Final Output.** (Proceed to generate output based on step 6).

1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Input:** Three documentation diffs (`hooks.md`, `interactive-mode.md`, `sub-agents.md`). * **Output Requirement:** Concise, insightful analysis in Markdown format. * **Language:** Chinese. * **Structure:** 1. Overall Summary (1-2 sentences). 2. Key Themes. 3. Impact Level with reasoning. 4. Action Items. 2. **Analyze the Diffs:** * **Diff 1: `hooks.md`** * *Change:* Added a sentence to the `TaskCreated` hook description: "In a [session without the Task tools](/docs/en/tools-reference#task-tool-availability), this event doesn't fire." * *Context:* The `TaskCreated` hook relies on `TaskCreate` tool usage. * **Diff 2: `interactive-mode.md`** * *Change:* Added a paragraph about the Task List view behavior on newer models (Opus 4.8, Sonnet 5, Fable 5, Mythos 5). * *Context:* Newer models manage multi-step work internally without explicit "Todo" tools, so the task list stays empty by default unless opted in via `CLAUDE_CODE_ENABLE_TODO_TOOLS=1`. * *Reference:* Links to "Task tool availability". * **Diff 3: `sub-agents.md`** * *Change:* Added a sentence regarding subagents and task tools. * *Context:* In a session without Task tools, subagents *also* don't get task tools, even if they run a different model. It distinguishes between in-process teammates and split-pane teammates. 3. **Synthesize the Findings:** * **Overall Summary:** These changes clarify the behavior of the "Task" tools (TaskCreate, TaskList, etc.) and related features across different models and contexts. Specifically, they document that newer models (like Sonnet 5) operate *without* these tools by default, affecting hooks, the UI task list, and subagent capabilities. * **Key Themes:** * **Model-specific Tool Availability:** The introduction of new model families (Opus 4.8, Sonnet 5, etc.) that handle tasks internally without exposing explicit Task tools by default. * **Feature Dependency:** Several features (Hooks, Task List UI, Subagent capabilities) depend on the presence of Task tools and are disabled or modified if those tools aren't available. * **Opt-in Mechanism:** A clear path to restore the old behavior using `CLAUDE_CODE_ENABLE_TODO_TOOLS=1`. 4. **Determine Impact Level:** * *Impact:* **Medium**. * *Reasoning:* This isn't a "breaking change" in the traditional API sense (code won't crash), but it changes the *behavior* developers might expect. If a developer upgrades Claude Code to use Sonnet 5, their hooks relying on `TaskCreated` might stop firing silently, and their custom UI might look empty. Developers building complex agent workflows need to know this nuance. 5. **Formulate Action Items:** * Check if your workflows rely on `TaskCreated` hooks. * If you prefer the explicit task list UI on newer models, set the environment variable. * Understand that subagents inherit the current session's tool constraints regarding Task tools. * Review the "Task tool availability" documentation for the full list of affected models. 6. **Draft the Response (in Chinese):** * *Summary:* 此次文档更新阐明了新模型（如 Sonnet 5）默认移除了 Task 工具，进而影响了相关钩子、UI 任务列表以及子代理的行为。 * *Themes:* 模型差异、功能依赖、显式启用。 * *Impact:* Medium (中等)。因为行为发生了重大变化，虽然是“默认”改变，但如果不注意，依赖这些功能的高级用户会感到困惑。 * *Actions:* 设置环境变量、检查钩子、了解子代理限制。 7. **Refine the Language (Professional & Concise):** * *Overall Summary:* 文档更新阐明了新模型版本（如 Sonnet 5、Opus 4.8）默认不再启用 Task 工具，这一改变影响了 `TaskCreated` 钩子的触发、UI 任务列表的显示以及子代理的工具继承行为。 * *Key Themes:* * **模型行为变更**：新模型通过内部逻辑而非显式工具管理多步骤工作，导致相关工具默认不可用。 * **功能连锁反应**：Task 工具的缺失导致 `TaskCreated` 钩子不触发、任务列表为空、子代理无法获取 Task 工具。 * **可回退机制**：提供了环境变量（如 `CLAUDE_CODE_ENABLE_TODO_TOOLS=1`）用于恢复旧版显式工具行为。 * *Impact:* **中等**。虽然这主要是行为上的变化而非语法破坏，但若开发者依赖 `TaskCreated` 钩子或可视化的任务列表来追踪工作流，升级模型后将面临功能“静默失效”的风险。 * *Action Items:* * 如果依赖可视化的任务列表，请设置 `CLAUDE_CODE_ENABLE_TODO_TOOLS=1` 环境变量。 * 检查现有的 `TaskCreated` 钩子逻辑，确认其是否需要适配新模型的默认行为。 * 注意子代理在无 Task 工具的会话中也无法继承这些工具，即使它们运行的是旧模型。 8. **Final Review:** Does this match the prompt's constraints? Yes. World-class programmer tone, markdown, Chinese, specific sections. * *Self-Correction during drafting:* Ensure I mention that subagents running in split panes might behave differently (as per the diff "runs as a separate Claude Code process, so its own model decides"), but the main point is about the *current session's* context affecting in-process agents or inherited tools. The diff says "In a session without the Task tools... Claude Code doesn't provide the task tools to subagents either". I should keep it simple: subagents follow the session's rules. 9. **Final Output Generation.** (Proceeds to generate output based on step 7).