不是工具清单，而是一套验证过程Not a tool list, but a validation process
这不是一份工具清单，也不被包装成已经成熟可靠的“个人操作系统”。它是一套持续运行的个人实验：把长期记忆、Obsidian 知识库、每日复盘、定时信息流和实战经验库接进日常，再把每次失败变成下一条可复用规则。
This is not a tool list or a claim of a finished, fully reliable “personal operating system.” It is an ongoing personal experiment: bringing long-term memory, Obsidian, daily reviews, scheduled information flows, and an experience library into real life, then turning each failure into a reusable rule.
- 发现真实问题Find a real problem从信息过载、上下文丢失、提醒失效或重复劳动中找到具体阻力。Identify concrete friction in information overload, lost context, failed reminders, or repeated work.
- 搭建并接入日常Build and use it daily让 Agent、知识库或自动化真正参与任务，而不是停留在演示。Put agents, knowledge, or automation into real tasks instead of leaving them as demos.
- 暴露不可靠Expose unreliability接受系统出现过期数据、重复投递、权限不可达和模型故障。Let stale data, duplicate delivery, permission gaps, and model failures become visible.
- 回到事实源Return to the source重新核验状态，修正判定规则，而不是接受一个听起来合理的答案。Recheck current state and fix the rule rather than accepting an answer that merely sounds plausible.
- 沉淀或关停Keep it or remove it有效经验进入知识库，失去价值的提醒和自动化被删除，重要控制权回到人手里。Useful lessons enter the knowledge base; low-value reminders and automations are removed, and important control returns to the human.
从问题到长期使用From problem to sustained use
我把信息过载、上下文丢失、提醒失效和重复劳动视为需要解决的问题，而不是先堆砌工具。通过长期记忆、知识库、复盘和按需自动化，让真实任务进入系统；每次异常后都回到事实源，修正规则。
I treat information overload, lost context, failed reminders, and repeated work as problems to solve—not reasons to accumulate more tools. Long-term memory, knowledge, reviews, and selective automation are used in real tasks; after each failure, I return to the source of truth and revise the rule.
这里不以功能数量或自动化数量作为结果，而以是否减少重复劳动、保留可追溯事实，并让重要决定仍由人掌控来评估。没有持续价值的流程会被关停。
Here I do not measure success by feature or automation count. I evaluate whether the system reduces repeated work, preserves traceable facts, and keeps consequential decisions under human control. Workflows without sustained value are shut down.
目前持续维护的四类能力Four capabilities under active development
- 长期记忆Long-term memory用 Markdown 作为事实源，检索索引用于定位，避免让缓存替代真实记录。Markdown remains the source of truth, with search indexes used for retrieval rather than replacing the record.
- 知识库Knowledge base使用 Obsidian 与 Agent 连接任务、项目、学习资料和可复用经验。Obsidian and agents connect tasks, projects, learning material, and reusable experience.
- 复盘系统Review system要求复盘读取真实日程、任务和记录，压缩输出，只保留能改变下一步行动的结论。Reviews use real schedules, tasks, and records, then compress the output to conclusions that can change the next action.
- 定时信息流Scheduled information源码更新、个性化摘要与记忆整理按需运行，无效或打扰过多的流程会被主动关停。Source updates, personalized summaries, and memory organization run when useful; noisy or low-value flows are deliberately shut down.
我保留的判断The judgment I keep
AI 应该降低认知负担，而不是制造更多需要维护的自动化。它可以承担检索、整理和重复执行，但事实核验、优先级和重要决策仍需要人负责。
AI should reduce cognitive load, not create more automation to maintain. It can retrieve, organize, and repeat, but factual validation, priorities, and consequential decisions still need human ownership.