Touchskyer's Thinking Wall — tech notes and essays.https://touchskyer.me/https://touchskyer.me/books/silicon-team/s3/ep08-trust-ledger--en/https://touchskyer.me/books/silicon-team/s3/ep08-trust-ledger--en/Three-season lookback: S1 established trust, S2 stress-tested trust, S3 extended trust. 163 stars is a number; the five-layer ledger is a structure. A tool's lifecycle isn't build → ship → done — it's build → ship → people arrive → discover trust only transferred one layer → keep fixing.Mon, 08 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep08-trust-ledger/https://touchskyer.me/books/silicon-team/s3/ep08-trust-ledger/三季回看：S1 建立信任，S2 压力测试信任，S3 扩展信任。163 stars 是一个数字，五层账本是一个结构。工具的生命周期不是 build → ship → done，是 build → ship → 有人来了 → 发现信任只转移了一层 → 继续修。Mon, 08 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep07-governance-layer--en/https://touchskyer.me/books/silicon-team/s3/ep07-governance-layer--en/Five role PRs with no unified review standard — who decides when a role is 'good enough'? Accepting contributions doesn't equal the ability to govern contributions. Open source without governance isn't an ecosystem — it's accumulation.Sun, 07 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep07-governance-layer/https://touchskyer.me/books/silicon-team/s3/ep07-governance-layer/五个角色 PR 没有统一的 review 标准——谁来判断一个角色'够好'？接受贡献不等于有能力治理贡献。没有治理的开源不是生态，是堆积。Sun, 07 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep06-fail-path-verified--en/https://touchskyer.me/books/silicon-team/s3/ep06-fail-path-verified--en/S1-S2's biggest unpaid technical debt: eight products, dozens of gates, the loop mechanism never triggered once. A safety net must be tested to count as a safety net — if you don't test it, nobody else will trust it.Sat, 06 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep06-fail-path-verified/https://touchskyer.me/books/silicon-team/s3/ep06-fail-path-verified/S1-S2 最大的未偿还技术债：八个产品，几十次门禁，回环机制一次都没触发过。安全网必须被测试过才算安全网——你自己不测，别人更不会信。Sat, 06 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep05-make-it-runnable--en/https://touchskyer.me/books/silicon-team/s3/ep05-make-it-runnable--en/One filename's capitalization rippled across 11 files. But #21 was just the tip of the iceberg — hardcoded paths, missing config layers, no session isolation, manual crash recovery. Making a personal tool public isn't polish — it's rebuilding the foundation of trust.Fri, 05 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep05-make-it-runnable/https://touchskyer.me/books/silicon-team/s3/ep05-make-it-runnable/一个文件名的大小写牵动了 11 个文件。但 #21 只是冰山一角——路径硬编码、配置分层缺失、会话（session）隔离不存在、崩溃恢复靠手动。把个人工具改成公共工具，不是 polish，是重建信任的地基。Fri, 05 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep04-core-not-public-asset--en/https://touchskyer.me/books/silicon-team/s3/ep04-core-not-public-asset--en/28 PRs, 0 external contributors touching harness/gate/review flow. They're not timid — the core never issued an invitation to change it. Public code doesn't mean public understanding — tradeoffs, abandoned alternatives, and historical bug context are all locked in the author's head.Thu, 04 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep04-core-not-public-asset/https://touchskyer.me/books/silicon-team/s3/ep04-core-not-public-asset/28 个 PR，0 个外部贡献者碰 harness/gate/review flow。不是他们胆小，是核心没有发出'可改'的邀请。代码公开不等于理解公开——权衡、放弃的方案、历史 bug 的上下文全锁在作者脑子里。Thu, 04 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep03-why-roles-work/https://touchskyer.me/books/silicon-team/s3/ep03-why-roles-work/五个角色 PR 都遵循了四段式格式。不是因为文档写得好，而是因为接口设计暗含了三个信号：看得懂、改得动、改坏了不怕。低风险贡献面是信任转移的入口。Wed, 03 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep03-why-roles-work--en/https://touchskyer.me/books/silicon-team/s3/ep03-why-roles-work--en/All five role PRs followed the four-section format. Not because documentation was good, but because the interface design implicitly signaled three things: understandable, modifiable, and safe to break. Low-risk contribution surfaces are the entry point for trust transfer.Wed, 03 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep02-they-add-roles-not-core/https://touchskyer.me/books/silicon-team/s3/ep02-they-add-roles-not-core/五个 PR，五个新角色——Performance、Technical Writer、Localization、Data Engineer、SRE。全部只改 roles/*.md，全部遵循四段式格式。零个 PR 碰 harness、gate 或 review flow。信任先转移到最安全的接口。Tue, 02 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep02-they-add-roles-not-core--en/https://touchskyer.me/books/silicon-team/s3/ep02-they-add-roles-not-core--en/Five PRs, five new roles — Performance, Technical Writer, Localization, Data Engineer, SRE. All only modify roles/*.md, all follow the four-section format. Zero PRs touch harness, gate, or review flow. Trust transfers to the safest interface first.Tue, 02 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep01-stuck-at-the-door--en/https://touchskyer.me/books/silicon-team/s3/ep01-stuck-at-the-door--en/163 stars, 39 forks. Then the first truly valuable feedback arrived — not praise, but: 'It doesn't run on Linux.' The distance between skill.md and SKILL.md is the distance between 'I can use it' and 'others can use it.'Mon, 01 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s3/ep01-stuck-at-the-door/https://touchskyer.me/books/silicon-team/s3/ep01-stuck-at-the-door/163 stars，39 forks。然后第一个真正有价值的反馈来了——不是夸奖，而是：Linux 上跑不起来。skill.md 和 SKILL.md 之间的距离，就是'我能用'和'别人也能用'之间的距离。Mon, 01 Jun 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep08-three-layer-nesting/https://touchskyer.me/books/silicon-team/s2/ep08-three-layer-nesting/用 OPC 审 OPC 自己，14 个节点全 PASS——FAIL 路径从未被触发。一个从未行使过的强制机制，和一个不存在的强制机制，没有区别。Fri, 29 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep08-three-layer-nesting--en/https://touchskyer.me/books/silicon-team/s2/ep08-three-layer-nesting--en/Used OPC to audit OPC itself; all 14 nodes PASS — the FAIL path was never triggered. An enforcement mechanism that has never been exercised is indistinguishable from one that doesn't exist.Fri, 29 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep07-opc-viewer/https://touchskyer.me/books/silicon-team/s2/ep07-opc-viewer/JSON 日志没人看，审查报告无人读。opc-viewer 把 AI 团队的审查过程变成对话回放——16 个虚拟团队成员有名字、有头像、有颜色。看不到过程，就没法信结果。Thu, 28 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep07-opc-viewer--en/https://touchskyer.me/books/silicon-team/s2/ep07-opc-viewer--en/Nobody reads JSON logs. Nobody reads review reports. opc-viewer turned the AI team's review process into a conversation replay — 16 virtual team members with names, avatars, and colors. If you can't see the process, you can't trust the result.Thu, 28 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep06-design-intelligence--en/https://touchskyer.me/books/silicon-team/s2/ep06-design-intelligence--en/After the education tool's frontend refactor, two blues differed by 7 degrees of hue. Security, backend, and frontend reviewers all PASS. Design Intelligence gave the pipeline a palette guardrail, 28 AI smell detectors, and a reference layer — teaching machines to see design.Wed, 27 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep06-design-intelligence/https://touchskyer.me/books/silicon-team/s2/ep06-design-intelligence/教育工具前端重构后，两个蓝色相差 7 度色相，安全、后端、前端三个审查者全部 PASS。Design Intelligence 给流水线装了调色板护栏、28 个 AI 嗅探器和参考层——让机器看得见设计。Wed, 27 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep05-extension-evolution--en/https://touchskyer.me/books/silicon-team/s2/ep05-extension-evolution--en/After two products, core code grew from 800 to 2,000 lines. Capability contracts let extensions and core ignore each other, five hook types cut into five pipeline phases, and a circuit breaker ensures one bad extension can't kill the line.Tue, 26 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep05-extension-evolution/https://touchskyer.me/books/silicon-team/s2/ep05-extension-evolution/两个产品做下来，核心代码从 800 行胖到 2000 行。能力合约让扩展和核心互不认识，五种钩子切入流水线的五个阶段，熔断器保证一个坏扩展不会拖死整条线。Tue, 26 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep04-looking-outward/https://touchskyer.me/books/silicon-team/s2/ep04-looking-outward/30 个开源 AI 项目拆开看：上帝文件三基因、插件困难三角、标准化代价、测试密度不等于成功。每个发现都回扣了一个设计决策。先看别人踩过的坑，再决定自己踩哪个。Mon, 25 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep04-looking-outward--en/https://touchskyer.me/books/silicon-team/s2/ep04-looking-outward--en/30 open-source AI projects dissected: God File three genes, the plugin difficulty triangle, the cost of standardization, and test density ≠ success. Each finding calibrated a design decision. Look at others' pitfalls before choosing which ones to step in yourself.Mon, 25 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep03-extension-system/https://touchskyer.me/books/silicon-team/s2/ep03-extension-system/四个审查者全判 PASS，没有一个质疑方向。加了第十人之后，第一次审查就发现了「想象力的失败」。共识不是推进的信号，是需要挑战的信号。Sun, 24 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep03-extension-system--en/https://touchskyer.me/books/silicon-team/s2/ep03-extension-system--en/Four reviewers unanimously judged PASS — not one questioned direction. After adding the Tenth Man, the very first review uncovered 'a failure of imagination.' Consensus isn't a signal to proceed — it's a signal to challenge.Sun, 24 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep02-pi-math-40-ticks/https://touchskyer.me/books/silicon-team/s2/ep02-pi-math-40-ticks/47 小时，$347，3 次上下文溢出。把一个散乱的单用户工具推到云端生产环境——五次部署调试，以及一个跑完了不肯停的循环。Sat, 23 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep02-pi-math-40-ticks--en/https://touchskyer.me/books/silicon-team/s2/ep02-pi-math-40-ticks--en/47 hours, $347, 3 context blowouts. Turning a single-user tool into SaaS — and the most ironic bug was the AI that wouldn't stop.Sat, 23 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep07-when-tools-check-themselves/https://touchskyer.me/books/silicon-team/s1/ep07-when-tools-check-themselves/用 OPC 审计 OPC，所有 gate 都 PASS——这才是最大的问题。一个关于镜子、洋葱和盲点的故事。Fri, 22 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep07-when-tools-check-themselves--en/https://touchskyer.me/books/silicon-team/s1/ep07-when-tools-check-themselves--en/Using OPC to audit OPC, every gate PASS — that's the real problem. A story about mirrors, onions, and blind spots.Fri, 22 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep01-family-calendar-23-hours-92-dollars/https://touchskyer.me/books/silicon-team/s2/ep01-family-calendar-23-hours-92-dollars/S1E10 扳回了方向，但代码还是那套 Tailwind 管理后台。LLM Pool 三路回退、Apple 设计令牌、WCAG 对比度——从方向正确到产品可用之间，是密度最高的打磨。Fri, 22 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s2/ep01-family-calendar-23-hours-92-dollars--en/https://touchskyer.me/books/silicon-team/s2/ep01-family-calendar-23-hours-92-dollars--en/One-sentence requirement, 44 AI subagents, 35 commits. Direction completely wrong midway, corrected by a single sentence. The two-day journey from zero to MVP.Fri, 22 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep10-when-humans-should-intervene/https://touchskyer.me/books/silicon-team/s1/ep10-when-humans-should-intervene/AI 跑了 23 小时方向全错，一句话就扳回来了。autonomous loop 不是不需要人，而是需要人在对的时间说对的话。Thu, 21 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep11-after-the-crash/https://touchskyer.me/books/silicon-team/s1/ep11-after-the-crash/Gate 全绿但封面图缩了 40%。review 系统 review 自己时崩了。当安全网本身有洞，修复不是回滚——是让系统学到教训。S1 终章。Thu, 21 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep11-after-the-crash--en/https://touchskyer.me/books/silicon-team/s1/ep11-after-the-crash--en/Gates all green, but the cover image shrunk 40%. The review system crashed when reviewing itself. When safety nets have holes, recovery isn't rollback—it's making the system learn. S1 finale.Thu, 21 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep10-when-humans-should-intervene--en/https://touchskyer.me/books/silicon-team/s1/ep10-when-humans-should-intervene--en/The AI ran 23 hours in the wrong direction. One sentence fixed it. Autonomous loops don't eliminate humans—they redefine when humans matter.Thu, 21 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep09-make-bad-outcomes-smaller--en/https://touchskyer.me/books/silicon-team/s1/ep09-make-bad-outcomes-smaller--en/One principle behind every OPC design decision: don't optimize AI's capability, optimize the consequences when it fails. Gates are floors, not sources. AI can make relative judgments, not absolute ones.Wed, 20 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep09-make-bad-outcomes-smaller/https://touchskyer.me/books/silicon-team/s1/ep09-make-bad-outcomes-smaller/OPC 所有设计决策背后的同一个原则：不要优化 AI 的能力，要优化它出错时的后果。Gate 是底线不是来源，AI 能做相对判断不能做绝对评价。Wed, 20 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep08-agent-amnesia--en/https://touchskyer.me/books/silicon-team/s1/ep08-agent-amnesia--en/By tick 8, AI started repeating work. It forgot it had just fixed this bug. Because its memory got compressed away.Tue, 19 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep08-agent-amnesia/https://touchskyer.me/books/silicon-team/s1/ep08-agent-amnesia/跑到第 8 个 tick，AI 开始重复工作。它忘了自己刚修过这个 bug。因为它的记忆被压缩掉了。Tue, 19 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep06-92-dollars-for-a-product--en/https://touchskyer.me/books/silicon-team/s1/ep06-92-dollars-for-a-product--en/23 hours, $92, 44 AI employees — the real bill from zero to a complete product. And why loops execute plans, they don't generate them.Sun, 17 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep06-92-dollars-for-a-product/https://touchskyer.me/books/silicon-team/s1/ep06-92-dollars-for-a-product/23 小时、$92、44 个 AI 员工——从零到一个完整产品的真实账单。以及为什么 Loop 执行计划，不生成计划。Sun, 17 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep05-ai-works-while-you-sleep/https://touchskyer.me/books/silicon-team/s1/ep05-ai-works-while-you-sleep/我跟 AI 说「跑完再来找我」，然后去睡觉了。几个小时后回来，一句话就让分数从 0.487 跳到了 0.68——不是因为 AI 做得好。Sat, 16 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep05-ai-works-while-you-sleep--en/https://touchskyer.me/books/silicon-team/s1/ep05-ai-works-while-you-sleep--en/I told AI 'finish up and come find me,' then went to bed. Hours later I checked back — one sentence pushed the score from 0.487 to 0.68. Not because AI did great work.Sat, 16 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep04-growing-a-skeleton--en/https://touchskyer.me/books/silicon-team/s1/ep04-growing-a-skeleton--en/From hardcoded to capability contracts — OPC's extension system design, and why 'joints are extension points, you can't hang muscles in the middle of a bone.'Fri, 15 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep04-growing-a-skeleton/https://touchskyer.me/books/silicon-team/s1/ep04-growing-a-skeleton/从 hardcoded 到 capability contract——OPC 的扩展系统设计，以及为什么「关节是扩展点，骨头中间不能挂肌肉」。Fri, 15 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep03-from-works-to-trusted--en/https://touchskyer.me/books/silicon-team/s1/ep03-from-works-to-trusted--en/Security audit score: 47. Command injection, path traversal, empty file cheating — after three weeks of hardening, OPC learned one thing: don't make AI better, make bad outcomes smaller.Thu, 14 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep03-from-works-to-trusted/https://touchskyer.me/books/silicon-team/s1/ep03-from-works-to-trusted/安全审计 47 分，命令注入、路径穿越、空文件作弊——三周硬化之后，OPC 学会了一件事：不让 AI 变好，让坏后果变小。Thu, 14 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep02-one-person-engineering-team/https://touchskyer.me/books/silicon-team/s1/ep02-one-person-engineering-team/4 种角色、14 个节点、1686 行机械规范——OPC 把「让 AI 帮我看看」变成了一条有状态的流水线。Wed, 13 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep02-one-person-engineering-team--en/https://touchskyer.me/books/silicon-team/s1/ep02-one-person-engineering-team--en/4 roles, 14 nodes, 1686 lines of mechanical spec — OPC turns 'let AI take a look' into a stateful pipeline.Wed, 13 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep01-why-ai-cant-engineer--en/https://touchskyer.me/books/silicon-team/s1/ep01-why-ai-cant-engineer--en/AI is a blazingly fast intern — but it doesn't understand what engineering means. The OPC story begins with this contradiction.Tue, 12 May 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s1/ep01-why-ai-cant-engineer/https://touchskyer.me/books/silicon-team/s1/ep01-why-ai-cant-engineer/AI 是一个写代码极快的实习生——但它不懂什么是工程。OPC 的故事从这个矛盾开始。Tue, 12 May 2026 00:00:00 GMThttps://touchskyer.me/blog/skill-is-living-knowledge/https://touchskyer.me/blog/skill-is-living-knowledge/文档是 write-once-read-maybe-update-never。Skill 是 human-agent contract 的持久化形式——agent 写、人 curate、知识在循环中活起来。Skill marketplace 为什么必然失败，每个人为什么需要自己的 Skill library。Tue, 14 Apr 2026 00:00:00 GMThttps://touchskyer.me/blog/skill-is-living-knowledge--en/https://touchskyer.me/blog/skill-is-living-knowledge--en/Documentation is write-once-read-maybe-update-never. A Skill is a persisted human-agent contract — agents write them, humans curate them, and knowledge stays alive in the loop. Why Skill marketplaces will inevitably fail, and why everyone needs their own Skill library.Tue, 14 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep01-preface--en/https://touchskyer.me/books/silicon-team/s0/ep01-preface--en/How one person builds an engineering team with AI agentsSun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep01-preface/https://touchskyer.me/books/silicon-team/s0/ep01-preface/这本书讲的是一个人如何用 AI agent 构建一支工程团队Sun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep03-harness-native--en/https://touchskyer.me/books/silicon-team/s0/ep03-harness-native--en/Constraint system design — how mechanical gates replace probabilistic judgmentSun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep04-multi-silicon--en/https://touchskyer.me/books/silicon-team/s0/ep04-multi-silicon--en/Agent routing, Spawn vs Delegate, multi-agent orchestrationSun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep02-protocol-layer/https://touchskyer.me/books/silicon-team/s0/ep02-protocol-layer/CLI 如何解耦 memory 和 agent，以及三层分发模型的设计Sun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep02-protocol-layer--en/https://touchskyer.me/books/silicon-team/s0/ep02-protocol-layer--en/How CLI decouples memory from agents, and the three-layer distribution modelSun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep04-multi-silicon/https://touchskyer.me/books/silicon-team/s0/ep04-multi-silicon/Agent routing、Spawn vs Delegate、多 agent 编排Sun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep05-autonomous-operation--en/https://touchskyer.me/books/silicon-team/s0/ep05-autonomous-operation--en/The OPC pipeline, iterative review, and autonomous loopsSun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep03-harness-native/https://touchskyer.me/books/silicon-team/s0/ep03-harness-native/约束系统设计——Mechanical Gate 如何替代概率性判断Sun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep05-autonomous-operation/https://touchskyer.me/books/silicon-team/s0/ep05-autonomous-operation/OPC pipeline、iterative review、autonomous loopSun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep06-startup-in-agent-era--en/https://touchskyer.me/books/silicon-team/s0/ep06-startup-in-agent-era--en/From engineering to business — positioning, narrative, investment thesisSun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep06-startup-in-agent-era/https://touchskyer.me/books/silicon-team/s0/ep06-startup-in-agent-era/从工程视角转向商业视角——定位、叙事、投资论点Sun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep08-memory-to-book/https://touchskyer.me/books/silicon-team/s0/ep08-memory-to-book/227 张卡片如何变成一本书Sun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep08-memory-to-book--en/https://touchskyer.me/books/silicon-team/s0/ep08-memory-to-book--en/How 227 cards became a bookSun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep07-engineering-lessons--en/https://touchskyer.me/books/silicon-team/s0/ep07-engineering-lessons--en/17 hard-won engineering lessonsSun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/books/silicon-team/s0/ep07-engineering-lessons/https://touchskyer.me/books/silicon-team/s0/ep07-engineering-lessons/17 个用血换来的具体教训Sun, 12 Apr 2026 00:00:00 GMThttps://touchskyer.me/blog/harness-engineering-ii--en/https://touchskyer.me/blog/harness-engineering-ii--en/Prompt engineering hits a cliff in long-run tasks. Not because your prompts are bad — because attention decay, context rot, and drift accumulation are physics. Harness engineering is the complete answer.Wed, 08 Apr 2026 00:00:00 GMThttps://touchskyer.me/blog/harness-engineering-ii/https://touchskyer.me/blog/harness-engineering-ii/Prompt engineering 在 long-run task 面前断崖式失效。不是因为 prompt 写得不好，是因为 attention decay、context rot、drift accumulation 三重物理极限。Harness engineering 是完整的答案。Wed, 08 Apr 2026 00:00:00 GMThttps://touchskyer.me/blog/agent-era-sustained-evolution/https://touchskyer.me/blog/agent-era-sustained-evolution/从认知阶梯到能力×验证矩阵：一年之后，我对 Agent 时代的新思考Tue, 07 Apr 2026 00:00:00 GMThttps://touchskyer.me/blog/agent-era-sustained-evolution--en/https://touchskyer.me/blog/agent-era-sustained-evolution--en/From the Cognition Ladder to the Capability × Validation Matrix: New thoughts on the Agent era, one year laterTue, 07 Apr 2026 00:00:00 GMThttps://touchskyer.me/blog/harness-engineering--en/https://touchskyer.me/blog/harness-engineering--en/LLMs satisfice by default. Harness engineering constrains agent behavior with structure, not prompts. A four-layer architecture and how we landed it in OPC.Sun, 05 Apr 2026 00:00:00 GMThttps://touchskyer.me/blog/harness-engineering/https://touchskyer.me/blog/harness-engineering/LLM 的默认行为是 satisfice，不是 optimize。Harness engineering 用结构约束 agent 行为，而不是用 prompt 期望 agent 行为。一个四层架构的设计理念和 OPC 落地。Sun, 05 Apr 2026 00:00:00 GMThttps://touchskyer.me/projects/opc/https://touchskyer.me/projects/opc/One Person Company — a digraph-based task pipeline with independent multi-role evaluation for AI agents.Wed, 01 Apr 2026 00:00:00 GMThttps://touchskyer.me/blog/opc--en/https://touchskyer.me/blog/opc--en/I built a Claude Code skill that dispatches 11 specialist AI agents to review code from different perspectives. Here's an honest comparison with just asking Claude directly.Sat, 28 Mar 2026 00:00:00 GMThttps://touchskyer.me/blog/opc/https://touchskyer.me/blog/opc/我做了一个 Claude Code skill，用 11 个 AI 专家并行 review 代码。这是一个诚实的对比：OPC vs 直接问 Claude。Sat, 28 Mar 2026 00:00:00 GMThttps://touchskyer.me/blog/memex/https://touchskyer.me/blog/memex/AI agent 每次新 session 都从零开始。我用 Zettelkasten 方法构建了 memex —— 一个不需要 vector DB 的 agent 持久记忆系统，背后的设计哲学和技术取舍。Tue, 24 Mar 2026 00:00:00 GMThttps://touchskyer.me/blog/memex--en/https://touchskyer.me/blog/memex--en/AI agents start from scratch every session. I built memex — a Zettelkasten-based persistent memory system that needs no vector DB. Here's the design philosophy and trade-offs behind it.Tue, 24 Mar 2026 00:00:00 GMThttps://touchskyer.me/projects/memex/https://touchskyer.me/projects/memex/Zettelkasten-based persistent memory for AI coding agents. No vector DB — just markdown + git.Fri, 20 Mar 2026 00:00:00 GMThttps://touchskyer.me/blog/superpowers--en/https://touchskyer.me/blog/superpowers--en/Superpowers is a set of skill plugins for Claude Code. A deep dive into its design philosophy — from enforced workflows to Zero Trust architecture, to a meta-layer mechanism that uses AI's weaknesses to constrain AI.Tue, 10 Mar 2026 00:00:00 GMThttps://touchskyer.me/blog/superpowers/https://touchskyer.me/blog/superpowers/Superpowers 是一套 Claude Code 的 skill 插件。深入分析其设计哲学：从流程强制到 Zero Trust 架构，再到用 AI 弱点约束 AI 的元层机制。Tue, 10 Mar 2026 00:00:00 GMThttps://touchskyer.me/projects/supercrew/https://touchskyer.me/projects/supercrew/Superpowers skills, hooks, commands, and agents for Claude Code.Sun, 01 Mar 2026 00:00:00 GMT