content: add practitioner skill collections + curated IA refactor

build/build.mjs

@@ -188,13 +188,14 @@ async function buildPages() {
       ? (lang === 'zh' ? `最后更新：${meta.lastUpdated}` : `Last updated: ${meta.lastUpdated}`)
       : '';
 
+    const sectionHasIndex = existsSync(join(SITE.contentDir, section, `index.${lang}.md`));
     const pageHtml = render(pageTemplate, {
       title: meta.title || '', content: html, lang,
       lastUpdatedLabel,
       pairPath: meta.pair ? `${SITE.base}/${pairLang}/${rel}/` : '', pairLang,
       sectionTitle: section.charAt(0).toUpperCase() + section.slice(1),
       homePath: `${SITE.base}/${lang}/`,
-      sectionPath: `${SITE.base}/${lang}/`,
+      sectionPath: sectionHasIndex ? `${SITE.base}/${lang}/${section}/` : `${SITE.base}/${lang}/`,
       contributeUrl: `${SITE.repo}/blob/main/CONTRIBUTING.md`,
     });
     const fullHtml = render(baseTemplate, {
@@ -280,8 +281,10 @@ async function buildIndex(pages) {
 
     let listingsHtml = '';
     for (const [section, items] of Object.entries(sections)) {
+      const indexPage = items.find(p => p.rel === section);
+      const visibleItems = indexPage ? [indexPage] : items;
       listingsHtml += `<section class="section"><h2>${section}</h2><ul>`;
-      for (const item of items) {
+      for (const item of visibleItems) {
         listingsHtml += `<li><a href="${item.path}">${item.title || item.path}</a></li>`;
       }
       listingsHtml += '</ul></section>';

content/curated/entries/practitioner-addyosmani-agent-skills.en.md

@@ -0,0 +1,27 @@
+---
+title: "Addy Osmani: agent-skills"
+description: "Commentary on Addy Osmani's production-grade skill collection for AI coding agents — 20 skills organized across the Define → Plan → Build → Verify → Review → Ship lifecycle."
+lang: en
+pair: practitioner-addyosmani-agent-skills.zh.md
+lastUpdated: 2026-05-09
+status: published
+source: https://github.com/addyosmani/agent-skills
+---
+
+# Addy Osmani: agent-skills
+
+**Source**: [github.com/addyosmani/agent-skills](https://github.com/addyosmani/agent-skills)
+
+## Why This Collection
+
+Addy Osmani is a long-standing practitioner in performance and web engineering at Google; his skill collection reflects that lineage — process-driven workflows that encode disciplined engineering rather than prompt tricks. The repository is organized around the software lifecycle (Define, Plan, Build, Verify, Review, Ship) and includes anti-rationalization tables that explicitly counter the shortcuts agents tend to take.
+
+## What's Inside
+
+- 20 skills mapped to six lifecycle phases, including `spec-driven-development`, `test-driven-development`, `code-review-and-quality`, `security-and-hardening`, and `ci-cd-and-automation`
+- Agent personas (code reviewer, test engineer, security auditor) for role-specific behavior
+- Mandatory verification gates rather than suggestions
+
+## Value for Practitioners
+
+Closest in spirit to this handbook's "agent-ready environment" framing. If you are designing skills for a team and want a reference for phase-by-phase coverage, this is the most complete public example. Treat the lifecycle taxonomy as a checklist: if your own skills leave a phase uncovered, that's a gap worth naming.

content/curated/entries/practitioner-addyosmani-agent-skills.zh.md

@@ -0,0 +1,27 @@
+---
+title: "Addy Osmani：agent-skills"
+description: "对 Addy Osmani 面向 AI 编码 agent 的生产级技能集合的解读——20 个技能按照 Define → Plan → Build → Verify → Review → Ship 生命周期组织。"
+lang: zh
+pair: practitioner-addyosmani-agent-skills.en.md
+lastUpdated: 2026-05-09
+status: published
+source: https://github.com/addyosmani/agent-skills
+---
+
+# Addy Osmani：agent-skills
+
+**来源**：[github.com/addyosmani/agent-skills](https://github.com/addyosmani/agent-skills)
+
+## 入选理由
+
+Addy Osmani 是 Google 长期深耕性能与 Web 工程的实践者；他的技能集合延续了这条脉络——以流程驱动的工作流编码工程纪律，而非提示词技巧。仓库围绕软件生命周期组织（Define、Plan、Build、Verify、Review、Ship），并通过"反合理化表格"显式对抗 agent 倾向于走的捷径。
+
+## 内容概览
+
+- 20 个技能映射到六个生命周期阶段，包括 `spec-driven-development`、`test-driven-development`、`code-review-and-quality`、`security-and-hardening`、`ci-cd-and-automation`
+- Agent 人格（code reviewer、test engineer、security auditor）用于特定角色行为
+- 强制验证环节而非建议
+
+## 对实践者的价值
+
+最贴近本 handbook "agent-ready 环境"表述的一份集合。如果你在为团队设计技能并希望参考分阶段的覆盖范围，这是目前公开的最完整样本。把其生命周期分类当作清单使用：如果你自己的技能有某个阶段没被覆盖，那就是一个值得命名的缺口。

content/curated/entries/practitioner-karpathy-llm-wiki.en.md

@@ -0,0 +1,27 @@
+---
+title: "Andrej Karpathy: LLM Wiki"
+description: "Commentary on Karpathy's LLM Wiki gist — an architecture where an LLM incrementally builds a persistent, cross-linked markdown knowledge base instead of re-deriving answers at query time."
+lang: en
+pair: practitioner-karpathy-llm-wiki.zh.md
+lastUpdated: 2026-05-09
+status: published
+source: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f
+---
+
+# Andrej Karpathy: LLM Wiki
+
+**Source**: [gist.github.com/karpathy/442a6bf555914893e9891c11519de94f](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f)
+
+## Why This Gist
+
+Karpathy rarely publishes on new architectures without sparking a wave of implementations, and this gist is no exception — dozens of open-source projects have tried to operationalize the idea within weeks. The proposal itself is short but load-bearing: instead of traditional RAG chunk-retrieval, have an LLM incrementally maintain a persistent markdown wiki where synthesis, cross-references, and contradictions are already resolved.
+
+## Key Ideas
+
+- Three layers: immutable raw sources, an LLM-maintained wiki of interlinked markdown pages, and a schema file (CLAUDE.md / AGENTS.md) governing LLM behavior
+- Three operations: **ingest** (process new sources), **query** (answer and file good answers back into the wiki), **lint** (periodic health checks for contradictions and gaps)
+- The wiki is a compounding artifact, not a retrieval cache
+
+## Value for Practitioners
+
+This frames a different end state for agent memory than the prevailing "retrieve more chunks" direction — one worth considering before you invest heavily in vector infra. Read the gist first, then the community debate in the comments: provenance, staleness, and whether "wiki" is even the right term are all unresolved and worth forming your own view on.

content/curated/entries/practitioner-karpathy-llm-wiki.zh.md

@@ -0,0 +1,27 @@
+---
+title: "Andrej Karpathy：LLM Wiki"
+description: "对 Karpathy LLM Wiki gist 的解读——一种由 LLM 增量构建持久、互联 Markdown 知识库的架构，而非在查询时重新推导答案。"
+lang: zh
+pair: practitioner-karpathy-llm-wiki.en.md
+lastUpdated: 2026-05-09
+status: published
+source: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f
+---
+
+# Andrej Karpathy：LLM Wiki
+
+**来源**：[gist.github.com/karpathy/442a6bf555914893e9891c11519de94f](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f)
+
+## 入选理由
+
+Karpathy 发布新架构思路时很少不会引发一波实现浪潮，这篇 gist 也不例外——数周内就有几十个开源项目试图将其工程化。提案本身很短但有重量：不同于传统 RAG 的切片检索，让 LLM 增量维护一份持久的 Markdown wiki，让综合、交叉引用与矛盾消解提前完成，而非查询时重新推导。
+
+## 核心要点
+
+- 三层结构：不可变的原始来源、由 LLM 维护的互联 Markdown wiki 页面、以及控制 LLM 行为的 schema 文件（CLAUDE.md / AGENTS.md）
+- 三种操作：**ingest**（处理新来源）、**query**（回答并把好答案回填入 wiki）、**lint**（周期性检查矛盾与缺口）
+- Wiki 是累积性产物，而非检索缓存
+
+## 对实践者的价值
+
+这为 agent 记忆描绘了与主流"检索更多切片"方向不同的终局状态——在你重度投入向量基础设施之前值得先考虑一下。建议先读 gist 本身，再看评论区的社区讨论：来源追溯、陈旧性、以及 "wiki" 是否是合适的术语都尚未定论，值得自己形成判断。

content/curated/entries/practitioner-mattpocock-skills.en.md

@@ -0,0 +1,27 @@
+---
+title: "Matt Pocock: skills"
+description: "Commentary on Matt Pocock's skill collection — small, composable, model-agnostic skills explicitly positioned against 'vibe coding'."
+lang: en
+pair: practitioner-mattpocock-skills.zh.md
+lastUpdated: 2026-05-09
+status: published
+source: https://github.com/mattpocock/skills
+---
+
+# Matt Pocock: skills
+
+**Source**: [github.com/mattpocock/skills](https://github.com/mattpocock/skills)
+
+## Why This Collection
+
+Matt Pocock is best known in the TypeScript community for teaching the type system by building up from first principles; that same pedagogical instinct shows in these skills. They are small, composable, and model-agnostic — designed to be picked up individually and combined, not adopted wholesale. The collection is explicitly framed as an alternative to "vibe coding": each skill enforces a specific engineering discipline.
+
+## What's Inside
+
+- Engineering skills like `/grill-with-docs` (align understanding before building), `/tdd`, `/diagnose` (structured debugging loop), and `/improve-codebase-architecture`
+- Productivity skills including `/grill-me` (deep questioning before starting work) and `/caveman` (compress agent output to cut token usage)
+- `/write-a-skill` for authoring new skills in the same style
+
+## Value for Practitioners
+
+Useful as a counterweight to larger, lifecycle-shaped collections like Addy Osmani's. When you want a single skill to inject discipline at one point in your workflow — not a full framework — this is the model to study. `/caveman` in particular is a good demonstration that skills can target the conversation itself, not just the code.

content/curated/entries/practitioner-mattpocock-skills.zh.md

@@ -0,0 +1,27 @@
+---
+title: "Matt Pocock：skills"
+description: "对 Matt Pocock 技能集合的解读——小巧、可组合、与模型无关的技能，显式对抗 'vibe coding'。"
+lang: zh
+pair: practitioner-mattpocock-skills.en.md
+lastUpdated: 2026-05-09
+status: published
+source: https://github.com/mattpocock/skills
+---
+
+# Matt Pocock：skills
+
+**来源**：[github.com/mattpocock/skills](https://github.com/mattpocock/skills)
+
+## 入选理由
+
+Matt Pocock 在 TypeScript 社区以"从第一性原理构建类型系统"的教学方式著称；这套技能延续了同样的教学本能。它们小巧、可组合、与模型无关——设计为可单独拾取、互相组合，而非整体采用。集合被明确定位为 "vibe coding" 的替代方案：每个技能都强制某种特定的工程纪律。
+
+## 内容概览
+
+- 工程类技能如 `/grill-with-docs`（构建前先对齐理解）、`/tdd`、`/diagnose`（结构化调试循环）、`/improve-codebase-architecture`
+- 生产力类技能包括 `/grill-me`（开始工作前的深度追问）和 `/caveman`（压缩 agent 输出以降低 token 消耗）
+- `/write-a-skill` 用于以同样风格编写新技能
+
+## 对实践者的价值
+
+作为对 Addy Osmani 这类生命周期型大集合的对照物很有用。当你想要一个单点注入工程纪律的技能，而非一整套框架，这是值得研究的范式。`/caveman` 尤其是一个好示例——它说明技能可以瞄准对话本身，而不仅仅是代码。

content/curated/index.en.md

@@ -1,19 +1,42 @@
 ---
 title: "Curated Articles"
-description: "A curated collection of authoritative articles on AI agent engineering from official sources, with selection rationale and practitioner value analysis."
+description: "A curated collection of authoritative articles and practitioner skill collections on AI agent engineering, with selection rationale and practitioner value analysis."
 lang: en
 pair: index.zh.md
-lastUpdated: 2026-05-07
+lastUpdated: 2026-05-09
 status: published
 ---
 
 # Curated Articles
 
-A curated collection of authoritative articles on AI agent engineering from official sources. Each entry includes why it was selected and what value it provides to practitioners.
+A curated collection of authoritative writing on AI agent engineering. Two kinds of entries live here: first-party articles from vendors building agent platforms, and skill collections from individual practitioners whose work has shaped how the community builds with agents. Each entry includes why it was selected and what value it provides to practitioners.
 
 ## Selection Criteria
 
+### For official articles
+
 - From official, first-party sources only
 - Provides actionable engineering insights (not marketing)
 - Current and relevant to modern agent practices
 - Verified and accessible
+
+### For practitioner collections
+
+- Clear attribution to a named author with traceable background
+- Independent thinking, not a restatement of vendor documentation
+- Publicly maintained and still active, or a clearly dated snapshot of a formative idea
+- Adds a distinct lens on how to build with agents — not redundant with existing entries
+
+## Official Articles
+
+- [Anthropic: Effective Context Engineering for AI Agents](/en/curated/entries/anthropic-context-engineering/) — how to manage system prompts, tools, retrieval, and memory for reliable agents
+- [Manus: Context Engineering for AI Agents](/en/curated/entries/manus-context-engineering/) — engineering lessons from building and scaling an autonomous agent product
+- [OpenAI: A Practical Guide to Building Agents](/en/curated/entries/openai-building-agents/) — OpenAI's first-party playbook for agent architecture and tool design
+
+## Practitioner Skill Collections
+
+- [Andrej Karpathy: LLM Wiki](/en/curated/entries/practitioner-karpathy-llm-wiki/) — a persistent-artifact architecture for agent memory, as an alternative to chunk-retrieval RAG
+- [Addy Osmani: agent-skills](/en/curated/entries/practitioner-addyosmani-agent-skills/) — production-grade skills organized by software lifecycle phase
+- [Matt Pocock: skills](/en/curated/entries/practitioner-mattpocock-skills/) — small, composable, model-agnostic skills as a counterweight to lifecycle frameworks
+
+The handbook curates rather than duplicates. Follow the source links; these collections evolve faster than this page.

content/curated/index.zh.md

@@ -1,19 +1,42 @@
 ---
 title: "精选文章"
-description: "来自官方来源的 AI agent 工程权威文章精选集，包含入选理由及对实践者的价值分析。"
+description: "AI agent 工程权威文章与实践者技能集合精选，包含入选理由及对实践者的价值分析。"
 lang: zh
 pair: index.en.md
-lastUpdated: 2026-05-07
+lastUpdated: 2026-05-09
 status: published
 ---
 
 # 精选文章
 
-来自官方来源的 AI agent 工程权威文章精选集。每篇条目包含入选理由及其对实践者的价值。
+AI agent 工程权威写作的精选集。本页收录两类条目：构建 agent 平台的厂商发布的一手文章，以及在社区中塑造 agent 构建方式的个人实践者所发布的技能集合。每篇条目包含入选理由及其对实践者的价值。
 
 ## 入选标准
 
+### 针对官方文章
+
 - 仅来自官方一手来源
 - 提供可操作的工程洞见（非营销内容）
 - 当前且与现代 agent 实践相关
 - 已验证且可访问
+
+### 针对实践者集合
+
+- 有明确署名、可追溯背景的作者
+- 独立思考，而非对厂商文档的转述
+- 公开维护且仍在更新，或是一个标注了时间的、关于某个成型思路的快照
+- 为如何构建 agent 提供独特视角——不与已有条目重复
+
+## 官方文章
+
+- [Anthropic：面向 AI Agent 的有效上下文工程](/zh/curated/entries/anthropic-context-engineering/) ——如何管理系统提示、工具、检索和记忆以构建可靠的 agent
+- [Manus：面向 AI Agent 的上下文工程](/zh/curated/entries/manus-context-engineering/) ——构建并规模化自主 agent 产品的工程经验
+- [OpenAI：构建 Agent 的实用指南](/zh/curated/entries/openai-building-agents/) ——OpenAI 一手发布的 agent 架构与工具设计实战手册
+
+## 实践者技能集合
+
+- [Andrej Karpathy：LLM Wiki](/zh/curated/entries/practitioner-karpathy-llm-wiki/) ——一种面向 agent 记忆的持久化产物架构，作为切片检索 RAG 的替代思路
+- [Addy Osmani：agent-skills](/zh/curated/entries/practitioner-addyosmani-agent-skills/) ——按软件生命周期阶段组织的生产级技能集合
+- [Matt Pocock：skills](/zh/curated/entries/practitioner-mattpocock-skills/) ——小巧、可组合、与模型无关的技能，作为生命周期框架的对照物
+
+本 handbook 做 curate 而非 duplicate。请顺着来源链接深入阅读；这些集合的演化速度远快于本页。