Changes on 2026-08-27

1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Input:** A batch of 18 documentation pages with diffs. * **Output:** Concise, insightful markdown analysis (Chinese). * **Focus:** What matters to developers. * **Structure:** 1. Overall Summary (1-2 sentences). 2. Key Themes (bullets). 3. Impact Level (Low/Medium/High/Breaking) with reasoning. 4. Action Items (max 5). 2. **Analyze the Diffs (Page by Page):** * `cli-reference.md`: Added `--restricted` flag. This is a security/sandboxing feature for evaluation harnesses. Disables command/code tools, restricts file access to working dirs, and blocks `bypassPermissions`. * `costs.md`: Added details about `/usage` showing usage-credits spend for Pro, Max, Team, and Enterprise plans. * `hooks.md`: Updated note regarding `bypassPermissions` not working if started in `restricted` mode. * `hooks-guide.md`: Same as `hooks.md`. * `interactive-mode.md`: Updated `Ctrl+O` (transcript viewer) description to mention expanding collapsed lines like cross-session messages. * `mcp.md`: Significantly restructured the "Automatic reconnection" section. Clarifies behavior for HTTP/SSE vs. Stdio servers, mid-session drops vs. first connection failures, and how Claude reports failures (via ToolSearch). * `monitoring-usage.md`: Minor cleanup. Removed specific examples (Prometheus, ClickHouse) from backend descriptions, keeping generic terms. * `network-config.md`: Minor text change. Removed specific names "CrowdStrike Falcon and Zscaler" from the CA certificate store section, kept generic "Enterprise TLS-inspection proxies". * `output-styles.md`: Clarified that output styles are added to the system prompt. Clarified when reminders are triggered (only when non-default is selected). * `overview.md`: Cleanup. Removed repeated `theme={null}` attributes from code blocks. * `plugin-marketplaces.md`: Removed version requirement (`v2.1.152`) for `relevance`. Clarified `allowlist` URL matching. * `plugins-reference.md`: Added logic for loading plugin agents without valid frontmatter (names them after file). Added command `claude plugin validate` to find bad frontmatter. * `quickstart.md`: Cleanup. Removed repeated `theme={null}` attributes. * `sandboxing.md`: Significant updates on merging scopes and excluding settings sources (`--setting-sources`). Details on how `mask` and `deny` entries behave when sources are excluded. * `skills.md`: Updated section on "Skills from additional directories". Added conditions regarding `project` setting source, `--safe-mode`, `strictPluginOnlyCustomization`, and bare mode. Clarified argument handling (literal text insertion vs expansion). * `slash-commands.md`: Duplicate of the changes in `skills.md` (seems like shared content). * `sub-agents.md`: Added warning for agent description token limits (>15k). Updated `maxTurns` description to mention partial output marking and resumption. Updated best practices ("specific" instead of "detailed"). * `CHANGELOG.md.md`: Confirms version 2.1.248. Highlights: `--restricted` flag, `experimental.cacheTtl`, `/usage-credits`, various bug fixes (MCP reconnection, hooks, agents), cleanup of generic examples. 3. **Synthesize Key Themes:** * **Security & Isolation:** The new `--restricted` mode is the biggest feature addition. It's explicitly for safe evaluation environments. * **Configuration & Scoping:** Deep dive into how settings from different scopes (user, project, managed) interact, specifically regarding exclusions (`--setting-sources`), sandbox rules (`mask`/`deny`), and loading resources (skills/agents) from added directories. * **Observability & Operations:** Updates to usage tracking (`/usage`), MCP connection reliability/behavior details, and monitoring documentation cleanup. * **Refinement & Polish:** Removal of specific product names (Zscaler, ClickHouse) in docs, cleanup of code block attributes, and clarifications on agent/skill behavior (token limits, argument expansion). 4. **Determine Impact Level:** * **Impact:** Medium. * **Reasoning:** While `--restricted` is a significant new feature for CI/CD and automated evaluation harnesses, most general developer workflows remain unchanged. The changes to sandboxing and settings merging are nuanced refinements. There are no breaking changes to existing CLI commands or core workflows, but there are important behavioral clarifications regarding "restricted mode" and settings merging. 5. **Draft Action Items:** * *Action 1:* Use `--restricted` in CI/CD or shared machine evaluations to enforce safety. * *Action 2:* Review sandboxing config if using `--setting-sources` to ensure security rules apply as expected. * *Action 3:* Check agent descriptions if hitting token limits. * *Action 4:* Validate plugin frontmatters using the new command if developing plugins. * *Action 5:* Update scripts relying on specific docs examples if they referenced removed generic product names (low priority, mostly readability). 6. **Drafting the Content (in Chinese):** * *Summary:* This update primarily introduces a security-focused `--restricted` mode for evaluation environments, refines sandboxing and settings scoping rules, and clarifies MCP reconnection behavior. * *Themes:* * Security: New restricted mode for safe automation. * Configuration: Deepened control over settings scopes and sandbox rules. * Reliability: Improved documentation on MCP reconnection logic. * Polish: Cleanup of documentation code blocks and examples. * *Impact:* Medium. New flags and logic changes affect advanced usage but don't break core features. * *Action Items:* 1. Try `--restricted` for CI/CD. 2. Verify sandbox configs with `--setting-sources`. 3. Check agent description lengths. 4. Run `claude plugin validate` during plugin dev. 7. **Refining the Output:** * Make sure the distinction between `skills.md` and `slash-commands.md` changes is understood (shared content). * Highlight the specific behavior of `bypassPermissions` being ignored in restricted mode. * Note the `claude plugin validate` addition. 8. **Final Review against Constraints:** * Markdown format? Yes. * Chinese? Yes. * Concise? Yes. * Focus on developers? Yes. *(Self-Correction during drafting)*: The `sandboxing.md` changes about `mask` and `deny` entries being ignored or kept when excluding sources are quite technical and important for security. I should emphasize that. Also, the MCP changes are clarifications, not new features, but important for debugging connection issues. 9. **Final Output Generation.** (Proceed to generate markdown).

1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Input:** Two pages of diffs (`mcp.md` and `about-claude/pricing.md`). * **Output Format:** Markdown, concise, insightful, Chinese language. * **Structure:** 1. Overall Summary (1-2 sentences). 2. Key Themes (bullets). 3. Impact Level (Low/Medium/High/Breaking) with reasoning. 4. Action Items (bullets, max 5). 2. **Analyze Page 1: `mcp.md` (+18 / -5):** * *Diff Content:* * Clarification on status reporting: Statuses in `claude mcp list` are now clarified as configuration decisions rather than connection attempts (for specific states like Pending/Rejected/Disabled). * New Status: "Disabled for this project" explicitly mentioned. * Error Reporting: When authentication/connection fails, Claude Code now adds the HTTP status/error code and the *origin* (scheme, host, port) of the URL to the error message, but explicitly omits the path/query/credentials. * Configuration Warnings: Expanded list of warnings: * Hidden whitespace (existed, now part of a list). * Same name in multiple scopes (New warning). * Reserved names (existed, now part of a list). * Missing environment variable (New warning). * *Developer Relevance:* This is mostly about better debugging and clearer feedback when MCP (Model Context Protocol) servers fail to connect or are misconfigured. The new warnings about scope conflicts and missing env vars are helpful for troubleshooting. 3. **Analyze Page 2: `about-claude/pricing.md` (+4 / -4):** * *Diff Content:* * Changed FAQ headers from bold text (`**Heading**`) to Markdown H3 headers (`### Heading`). * *Developer Relevance:* Minimal functional impact. It's a documentation formatting consistency update. 4. **Synthesize the Analysis:** * *Overall Summary:* Documentation updates primarily improve the clarity of MCP (Model Context Protocol) server status reporting and error diagnostics, alongside minor formatting consistency updates in the pricing section. * *Key Themes:* * **Enhanced Diagnostics:** MCP error messages now include more context (origin of URL) without exposing secrets. * **Configuration Validation:** New warnings added for conflicting server names across scopes and missing environment variables. * **Status Semantics:** Clarified that list statuses reflect configuration states, not necessarily active connection health checks (in specific cases). * **Documentation Formatting:** Standardized heading levels in the Pricing FAQ. * *Impact Level:* **Low**. * *Reasoning:* These changes are documentation-centric, clarifying behavior and error reporting rather than breaking existing functionality. Developers benefit from better error messages but don't need to rewrite code. * *Action Items:* * Review any custom MCP configurations for duplicate names across scopes or missing environment variables. * Be aware that `claude mcp list` output for disabled/rejected servers has changed semantics (config status vs. connection check). * Use the improved error messages (URL origins) to debug connection failures more efficiently. 5. **Draft the Response (in Chinese):** * *Drafting Summary:* 文档更新主要集中在提升 MCP 服务器状态报告和错误诊断的清晰度，以及定价部分的格式规范化。 * *Drafting Key Themes:* * 错误诊断增强：MCP 连接失败时，错误信息现在会包含 URL 源，但会隐藏敏感路径。 * 配置验证：新增了针对多作用域重名服务器和缺失环境变量的警告。 * 状态语义：明确了列表状态反映的是配置状态而非连接尝试（针对特定情况）。 * 文档格式：定价 FAQ 标题标准化。 * *Drafting Impact:* Low. 主要是文档层面的改进，优化了错误反馈和警告信息，并未破坏现有功能。 * *Drafting Action Items:* * 检查 MCP 配置中是否存在跨作用域重名或环境变量缺失。 * 利用改进后的错误信息（包含 URL 源）来调试连接问题。 * 留意 `claude mcp list` 中针对禁用/拒绝服务器的状态描述变化。 6. **Refine and Polish:** * Ensure the tone is professional ("World-class programmer"). * Check constraints (max 5 action items). *Self-Correction during drafting:* The pricing change is trivial. I should focus heavily on the `mcp.md` changes because that's where the technical meat is. The "Same name in more than one scope" is a specific new warning that could cause confusion if not addressed, so it's worth highlighting. 7. **Final Output Generation:** (Proceed to generate the markdown output).