「Why Claude Code」缺一个可对照的参照物：学生搭完 nano harness 后无法验证真实 Claude Code 发了什么
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 2/5
- Estimated time
- 1-3 hours
- Newbie friendliness
- 85/100
- Issue type
- Documentation
- Clarity
- Clearly specified
- Activity status
- Active
- Domain
- documentation
Research direction
Start at the “Why Claude Code” section in README.md, README-zh.md, and README-ja.md, and review the surrounding wording and link style. Add the requested reference link at the end of that section in all three files; done means the links are present, point to the specified Claude-Code archive, and the chapter code and course structure are unchanged.
Written by the indexing model from the issue text.
Description
按 CONTRIBUTING 第 1 条，先开 issue 说清楚问题，再决定要不要发 PR。
问题
README 的「Why Claude Code」把 Claude Code 剥到本质，列了一串构件（agent loop + tools + skill 按需加载 + context compaction + subagent + task system + permission governance…）。17 课就是照着这张清单一节一节搭出来的。
但学生搭完之后，没有办法把自己的 nano harness 和真实的那一个对照——课程从头到尾没有出现过真实 Claude Code 实际发到模型的那段系统提示词和工具 schema。于是「Claude Code 是这么做的」这句话，学生只能信，不能验。
具体会卡在两处：
- s03_permission：真实 harness 的权限治理有多少是写在提示词里、多少是写在工具 schema 里？学生没有参照物，只能照课程的简化版理解。
- 入口差异：真实 Claude Code 从终端进和从 SDK 进，发的提示词和工具集并不一样——同模型同日，一边约 26k 字符 35 个工具，另一边约 21k 字符 29 个工具，连开头那句身份都改写了。这件事恰好印证课程反复强调的「Agent 产品 = 模型 + harness」：模型没变，harness 的入口变了，产品行为就变了。课程里没有地方能让学生看到这个。
想提的改动
在 README 的「Why Claude Code」那节末尾加一行参考链接，指向按产品分目录、带日期、带复现命令的系统提示词与工具 schema 实抓归档，让学生能自己跑一遍 diff：
一行，三语 README 同步（README.md / README-zh.md / README-ja.md），不动任何章节代码，不扩展课程结构。
两件要先说明白的
- 利益相关：这个归档由我所在的 Continuum AI 维护。所以我没有直接发 PR，先来问你们要不要——不要的话我就不发了，也不会再提。
- AI 协助：按 CONTRIBUTING 第 5 条声明，这条 issue 是我用 Claude Code 写的。上面两处数字（26k/35、21k/29）是我们自己从线上抓下来的实测，我核对过文件再写进来的，不是模型复述的。
另外看到 CONTRIBUTING 里「会直接关掉 agent 批量跨仓库开的 PR」那条——我这个账号最近确实在不少仓库提过东西，所以这条我主动摆出来：如果你们觉得这就是那一类，直接关掉就好，不用解释。
🤖 Generated with Claude Code
- Dominant language
- Python
- Stars
- 78.2k
- Forks
- 12.5k
- Avg merge
- 3h 27m
- Merged PRs (30d)
- 2
Getting set up
- No Dockerfile or Docker Compose file
- No pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from shareAI-lab/learn-claude-code
-
[web] Buttons show the default arrow cursor on hover (Tailwind v4 dropped the preflight `cursor: pointer`)Possibly taken @xbsheng claimed this 2 days ago. Open
Difficulty 2/5 1-3 hours Newbie friendliness 85/100
shareAI-lab/learn-claude-code#594 ·
Maintainers usually reply within 1 day
-
s04/s05/s06 的 README.zh.md 里有几处代码注释仍是英文Possibly taken @zl1nq claimed this 16 days ago. Open
Difficulty 2/5 1-3 hours Newbie friendliness 76/100
shareAI-lab/learn-claude-code#589 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
shareAI-lab/learn-claude-code#583 ·
Maintainers usually reply within 1 day
-
[s08] Reactive compaction does not recognize Anthropic's "prompt is too long" errorPossibly taken @zmylol claimed this 38 days ago. Open
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
shareAI-lab/learn-claude-code#560 · 1 comment ·
Maintainers usually reply within 1 day
-
docs(s02): explain why specialized tools are needed when Bash can do the same workPossibly taken @kunshanyep-beep claimed this 36 days ago. Open
Difficulty 1/5 Under an hour Newbie friendliness 76/100
shareAI-lab/learn-claude-code#549 · 1 comment ·
Maintainers usually reply within 1 day
All issues in shareAI-lab/learn-claude-code
Similar issues
-
md/asciidoc/csv: an explicit `encoding="utf-8"` keeps a leading BOM, the auto-detect path drops itOpen
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
docling-project/docling#4846 ·
Maintainers usually reply within 1 day
-
area/sessions comp/dashboard comp/desktop comp/gateway comp/tui duplicate P2 sweeper:risk-session-state type/bug
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
NousResearch/hermes-agent#138914 · 1 comment ·
Maintainers usually reply within 1 day
-
bot-triaged module: activation checkpointing module: inductor oncall: pt2 triaged
Difficulty 2/5 1-3 hours Newbie friendliness 62/100
Maintainers usually reply within 1 day
-
[oblt-aw][autodoc] Update PR Actions Detective E2E prerequisites and fix dashboard-format table typoOpendocumentation
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
elastic/oblt-aw#2382 · 1 comment ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
Maintainers usually reply within 1 day