SKILLS & AGENTS/Skill 与 Agent·MCP-SERVER

ai-agent-book

bojieli/ai-agent-book

《深入理解 AI Agent：设计原理与工程实践》（李博杰著）开源主仓库：全书正文、编译版 PDF 与按章配套代码

怎么装：Install: git clone https://github.com/anthropics/claude-quickstarts.git chapter6/claude-quickstarts && git -C chapter6/claude-quickstarts checkout --detach 9bcc95e316e5ef6542b4c9d0469f4078829eead5 # 实验 6-8 使用 computer-use-demo/

02 / 现在的位置02 / Why now

它现在处在什么位置Where it stands right now

首次发现FIRST SEEN

03 OCT 202603 OCT 2026

增长GROWTH

+0.2%+0.2%

52k → 52k stars · 2 个快照52k → 52k stars · 2 snapshots

状态STATUS

持续活跃Active

03 / 它能帮你做什么03 / What it helps you do

它能帮你做成什么What it helps you do

04 / 安装04 / Install

怎么装How to install

GIT-CLONE

$

README 里给出的安装方式。The installation method given in the README.

GIT-CLONE

$

README 里给出的安装方式。The installation method given in the README.

两种装法选一个即可 —— 上面那条给命令行用户，下面那条给写代码的用户。Pick one of the two — the first is for command-line users, the second for people writing code.

05 / 限制与风险05 / Limits and risk

限制与风险Limits and risk

风险不是警告，是可信度的一部分。以下结论只基于文档静态扫描，我们不会执行项目里的任何代码。Risk here is evidence, not an alarm. These findings come from static scanning of the docs; we never execute a project’s code.

限制LIMITS

需检查Needs review

01需要 API KeyREQUIRES API KEY: 是Yes
02需要本地环境NEEDS LOCAL RUNTIME: 是Yes
03需要 DockerNEEDS DOCKER: 未在文档中发现Not found in the docs
04难度DIFFICULTY: 需要开发经验Needs development work

PROVENANCE · 这些结论的依据是什么Where these conclusions come from

来源 URLSource URL: https://github.com/bojieli/ai-agent-book
原始数据哈希Raw data hash: 47330990bf3d04ec80f0cb15074376ca
解析时间Parsed at: 2026-10-04T09:07:17.646Z
最后一次验证Last verified: 2026-10-04T08:53:33.891Z
增长测量时间Growth measured: —
发现查询Discovery queries: focus:token:repo:topic-context-engineering
分类来源Category source: rules
焦点Focus: token · design (7)

静态扫描STATIC SCAN

01控制浏览器Controls a browserreadme

- 浏览器、CUDA、FFmpeg、Ollama、Playwright 浏览器及外部仓库等系统依赖，请继续参考各实验 README。第 8 章部分内置第三方组件需要 Python 3.12+。

02读写本地文件Reads and writes local filesskills/coding-agent-harness/SKILL.md

1. **工具集**：文件读写（read/write/edit）、目录浏览（glob/ls）、内容搜索（grep）、命令执行（bash）七件套；每个工具命名直观、参数带例子、边界有说明（防呆设计）。

03控制浏览器Controls a browserskills/computer-use/SKILL.md

- 没有编程 API 可用、Selenium 之类脚本化方案难以维护页面的场景。

这里只做静态扫描：读 README、SKILL.md 和依赖清单，不执行代码。没有命中不代表安全。This is static scanning only: we read the README, SKILL.md and dependency list, and never execute code. No findings does not mean safe.

完整解析证据 JSONFull parse evidence JSON

{
  "agents": [
    {
      "agent": "claude-code",
      "evidence": "readme: …ude 等 LLM 产品；熟悉至少一款 AI 辅助编程工具（Claude Code、Codex、Cursor 等）；了解命令行、Git、JSO…"
    },
    {
      "agent": "claude-desktop",
      "evidence": "skills/mcp-skill-hub/SKILL.md: …处处可用。** 一个 MCP 服务器可同时被 Cursor、Claude Desktop、OpenClaw 等任何兼容客户端使用，开发者无需关心上游…"
    },
    {
      "agent": "codex",
      "evidence": "skills/tool-discovery/SKILL.md: …置内联展开 block，官方文档明确后续每轮保持缓存命中）；Codex CLI 默认开启 `tool_search`。 - **真正导致重…"
    }
  ],
  "apiKey": "置第三方组件需要 Python 3.12+。 ## 🔑 API 密钥 建议申请下面几个平台的 API Key 方便学习。模型选型可参考 [这篇指南",
  "docker": null,
  "taxonomy": {
    "scores": {
      "design": 3,
      "ai-tools": 7,
      "dev-tools": 0,
      "indie-web": 0,
      "skill-agent": 3,
      "productivity": 3
    },
    "primary": "skill-agent",
    "secondary": "mcp-server"
  },
  "localRuntime": "环境变量。只有在实验 README 或 CLI 明确列出 `ollama` 时，才可启动本地 Ollama 并添加 `--provider ollama",
  "skillMdTotal": 30,
  "skillMdErrors": [
    {
      "path": "chapter8/curly-quote-sft/SKILL.md",
      "error": "no frontmatter"
    }
  ],
  "treeTruncated": false,
  "categoryScores": {
    "data": 0,
    "agent": 3,
    "media": 0,
    "design": 3,
    "browser": 0,
    "devtool": 0,
    "security": 0,
    "marketing": 0,
    "productivity": 3
  },
  "classification": [
    {
      "signal": "29 SKILL.md with name + description",
      "weight": 0.95
    },
    {
      "signal": "python dependency on mcp / fastmcp + MCP framing",
      "weight": 0.8
    },
    {
      "signal": "skills/ directory with 23 markdown files",
      "weight": 0.75
    },
    {
      "signal": "README describes a Claude / Agent Skill",
      "weight": 0.7
    },
    {
      "signal": "description mentions AI",
      "weight": 0.2
    }
  ],
  "scannedSources": [
    "readme",
    "chapter10/multi-role-transfer/skills/coding/SKILL.md",
    "chapter10/multi-role-transfer/skills/data_analysis/SKILL.md",
    "chapter10/multi-role-transfer/skills/research/SKILL.md",
    "chapter10/multi-role-transfer/skills/triage/SKILL.md",
    "chapter10/multi-role-transfer/skills/writing/SKILL.md",
    "chapter2/agent-skills-ppt/skills/pptx/SKILL.md",
    "chapter8/curly-quote-sft/SKILL.md",
    "chapter9/ai-style-skill/skill/SKILL.md",
    "skills/agent-evaluation/SKILL.md",
    "skills/agent-evolution/SKILL.md",
    "skills/agent-state-bar/SKILL.md",
    "skills/async-event-agent/SKILL.md",
    "skills/bad-case-to-dpo/SKILL.md",
    "skills/coding-agent-harness/SKILL.md",
    "skills/computer-use/SKILL.md",
    "skills/context-compression/SKILL.md",
    "skills/context-engineering/SKILL.md",
    "skills/error-recovery/SKILL.md",
    "skills/eval-dataset-design/SKILL.md",
    "skills/knowledge-org/SKILL.md",
    "skills/kv-cache-design/SKILL.md",
    "skills/loop-engineering/SKILL.md",
    "skills/mcp-skill-hub/SKILL.md",
    "skills/memory-system/SKILL.md",
    "skills/multi-agent-design/SKILL.md",
    "skills/post-training-strategy/SKILL.md",
    "skills/rag-pipeline/SKILL.md",
    "skills/reward-design/SKILL.md",
    "skills/tool-design/SKILL.md",
    "skills/tool-discovery/SKILL.md"
  ],
  "skillMdFetched": 30
}