docs(security/automation/settings): 完善 15 篇文档交叉引用和内容扩充 - #37
Conversation
复盘修正： - 移除之前不可验证的 AI 生成概念图（4张，各~5MB） - 移除捏造的翻译覆盖率数据（~95%） - 移除无出处的语言规划声明（日语、韩语在规划中） - 保留原始 i18n 路径（app/data/i18n/） - 未修改 04-compute-service.md（禁区） Security (5篇): - 01-controllability: 新增审批模式关联说明和相关页面导航 - 02-risk-levels: 新增相关页面导航 - 03-audit-trail: 新增相关页面导航 - 04-data-security: 新增相关页面导航 - 05-ai-approval: 扩展下一步链接 Automation (3篇): - 01-heartbeat: 新增相关页面导航 - 02-notifications: 审批通知段落新增到 ai-approval 的交叉引用 - 03-scheduled-tasks: 新增相关页面导航 Settings (7篇，跳过 compute-service): - 01-appearance: 扩充毛玻璃设计详解、主题影响范围、FAQ - 02-language: 扩充翻译回退机制、贡献流程、翻译规范 - 03-keyboard-shortcuts: 新增相关页面导航 - 05-permissions: 新增平台差异（macOS/Windows/Linux）和故障排除 - 06-data-privacy: 新增相关页面导航 - 07-user-profile: 新增用户画像与智能体记忆关系（基于 AgentFS 结构） - 08-about: 新增相关页面导航
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1b22a71e25
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".
|
|
||
| - **可见**：审批卡片展示操作类型、参数、来源追溯和风险等级 | ||
| - **可控**：你可以选择允许、拒绝或修改操作参数 | ||
| - **可逆**：即使批准了操作，仍可通过四级回滚撤销变更 |
There was a problem hiding this comment.
When an approved operation sends email, invokes an external API, pushes a remote commit, or changes files outside DesireCore's managed scope, the four rollback levels cannot undo that side effect; 02-conversations/10-rewind-checkpoints.md explicitly lists these limitations. Claiming here that an approved operation can still be rolled back may encourage users to approve irreversible actions under a false safety guarantee, so qualify this statement to cover only DesireCore-managed local changes.
Useful? React with 👍 / 👎.
| :::info | ||
| 画像信息完全存储在本地，你可以随时查看和编辑。删除画像文件后，智能体将不再参考历史信息，回到初始状态。 |
There was a problem hiding this comment.
When users/<ID>/agents/<agentID>/memory/ contains interaction history, deleting only profile.md does not return the agent to its initial state or stop it from referencing all historical information—the separate memory directory described immediately below remains intact. This overstates the privacy/deletion effect; explain that profile deletion removes only the shared profile and that per-agent memories must be cleared separately.
Useful? React with 👍 / 👎.
| ## 用户档案与智能体记忆 | ||
|
|
||
| 你的用户档案不仅是一个显示名称——它还是智能体了解你的基础。 |
There was a problem hiding this comment.
The site explicitly serves an en locale, and all 15 changed pages have dedicated mirrors under i18n/en/docusaurus-plugin-content-docs/current/, but this commit updates only the Chinese sources. Consequently English users continue receiving the old pages without the new settings guidance, profile-memory explanation, or navigation links; add equivalent updates to the English files.
Useful? React with 👍 / 👎.
|
|
||
| - [三层可控性](./01-controllability.md)：了解可见、可控、可逆的安全理念 | ||
| - [工具与命令审批模式](./05-ai-approval.md)：五种审批模式如何与风险等级配合 | ||
| - [权限设置](../10-settings/05-permissions.md)：管理系统权限和已记住的规则 |
There was a problem hiding this comment.
For users trying to manage a remembered approval rule, this link opens 05-permissions.md, which only documents operating-system permissions such as screen recording, microphone, and accessibility and contains no remembered-rule controls. Point the link to the agent approval-preferences documentation (or add the missing rule-management section) rather than describing the system-permissions page as managing remembered rules.
Useful? React with 👍 / 👎.
| DesireCore 的界面文案存储在 `app/data/i18n/` 目录下。如果你希望帮助改进翻译或贡献新语言，欢迎通过 GitHub 提交 Pull Request。 | ||
| DesireCore 的界面文案存储在 `app/data/i18n/` 目录下。如果你希望帮助改进翻译或贡献新语言： | ||
|
|
||
| 1. **改进现有翻译**：Fork [DesireCore 仓库](https://github.com/desirecore)，修改对应语言的翻译文件，提交 Pull Request |
There was a problem hiding this comment.
The contribution step tells readers to fork a repository, but https://github.com/desirecore is the organization landing page rather than a specific repository and therefore cannot itself be forked. Link directly to the repository that contains app/data/i18n/ so contributors can follow the documented workflow and edit the stated files.
Useful? React with 👍 / 👎.
- controllability-layers.png: 三层可控性递进概念图 - risk-levels-pyramid.png: 四级风险等级金字塔 - approval-flow.png: 审批决策流程图 - data-flow.png: 本地优先数据流向图 Each image is referenced from its corresponding markdown file.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 399c54c323
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".
|
|
||
| DesireCore 在设计之初就将数据安全作为核心考量。本页面帮助你了解你的数据是如何被保护的。 | ||
|
|
||
|  |
There was a problem hiding this comment.
The alt text says that only model requests leave the device, but this is false when the user signs in, launches the application, checks for updates, or invokes networked tools: this page itself documents startup statistics and update downloads, while 10-settings/06-data-privacy.md documents account-resource synchronization. Because screen readers and failed-image fallbacks present this sentence as the architecture description, it gives users an incorrect privacy guarantee; describe each outbound-data category rather than limiting traffic to model endpoints.
Useful? React with 👍 / 👎.
|
|
||
| DesireCore 为每个智能体提供独立的工具与命令审批偏好。审批发生在真正执行操作之前；提示词、Skill 或委派不能绕过这条执行边界。 | ||
|
|
||
|  |
There was a problem hiding this comment.
In the default AI-approval scenario, the new PNG routes Auto Execute into AI Risk Evaluation, while the approval-card node has no clear path into that evaluation. This contradicts the behavior documented immediately below: operations that require confirmation display the card while AI evaluates them, and the returned recommendation is used only when the 30-second timer expires. Redraw this branch so automatically permitted operations bypass evaluation and confirmation-required operations connect the card, timer, AI recommendation, and final decision correctly.
Useful? React with 👍 / 👎.
|
|
||
| DesireCore 的安全设计围绕一个核心理念：**你始终是最终决策者。** 智能体是你的助手，不是你的管理者。为了确保这一点，我们构建了三层可控性体系。 | ||
|
|
||
|  |
There was a problem hiding this comment.
Each of the four newly referenced 2048×2048 PNGs is 4.3–5.0 MB, whereas the largest pre-existing documentation image is under 0.8 MB. Since every diagram appears near the top of its page, visiting any affected security page immediately incurs a multi-megabyte image transfer, significantly slowing mobile and constrained-network users; downscale the artwork and encode it as an optimized WebP/AVIF or SVG asset.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ab5ee26164
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".
|
|
||
| - **你主动告知**：在对话中说"记住我叫张三，我是做后端开发的"，智能体会写入画像 | ||
| - **你手动编辑**：直接在设置中修改个人信息 | ||
| - **智能体学习**：在长期对话中，智能体可能识别到你的偏好并记录（需经你确认） |
There was a problem hiding this comment.
This tells users that learned preferences are recorded only after their confirmation, but docs/02-user-guide/07-memory/04-auto-learning.md:123-129 says the profile is extracted and updated automatically, and docs/04-concepts/06-self-evolution.md:36,57-60 explicitly marks automatic profile updates as requiring no confirmation. Users may therefore disclose preferences under an incorrect privacy expectation; document the actual automatic-update behavior and the available review/deletion controls instead.
Useful? React with 👍 / 👎.
1 participant
概述
对安全（Security）、自动化（Automation）和设置（Settings）三个章节共 15 篇文档进行完善，主要内容为交叉引用导航和已验证内容扩充。
复盘修正（相对之前直接 push main 的版本）
app/data/i18n/）04-compute-service.md（禁区）修改清单
Security (5篇)
Automation (3篇)
Settings (7篇，跳过 compute-service)
内容来源说明
所有新增内容均基于：