最难用的学习工具
翻譯準備中
技能Python · ShellGitHub @Vico304自行安裝營運者未確認
試用過了嗎? 點一下就好,不需要登入。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
安裝型專案無需部署網域也可驗證。個人儲存庫可透過 GitHub 登入驗證,組織儲存庫可使用上方的儲存庫驗證檔案。
正在載入完整 README。 以下為已儲存的摘錄。
Whetstone
A whetstone sharpens a blade. Blades don't get sharp by soaking in oil.
"I just love the feeling of knowledge flowing naturally into my brain." — said every clueless waste of a learner檢視完整 README ↗A whetstone sharpens a blade. Blades don't get sharp by soaking in oil.
"I just love the feeling of knowledge flowing naturally into my brain." — said every clueless waste of a learner
Why does everyone now believe they can actually learn from a short video? Lectures at 2x speed, papers "read" via a 100-word AI summary, agents driven with nothing but "continue", "it errored", "just give me something that runs". Somewhere along the way we forgot that learning is supposed to be a scourging of the soul, a grinding of the mind, a mortification of the flesh. If you want effortless, why not sleep on the textbook and absorb it by osmosis? :)
This is the most user-hostile learning tool there is. Unless you also want to be Sisyphus pushing the boulder with me, I don't recommend it to anyone.
This is the short English version. The full documentation is in Chinese: 中文说明 →. The skills teach in Chinese.
A learning tool that runs inside Claude Code, Claude Desktop, Codex, DeepSeek Harness or pi. Give it a book, a document or a codebase; it rearranges the material into a problem → solution → new problem path and then, section by section, makes you:
Decades of learning science point the same way: methods that make learning feel easy mostly hurt long-term retention; methods that make recall harder mostly strengthen it. Retrieval practice g ≈ 0.5–0.6, self-explanation g ≈ 0.55, rereading near zero (sources in docs/evidence.md). This tool sits on the effortful side.
Two things it deliberately withholds:
One skill directory, five hosts:
| Host | Install | Invoke |
|---|---|---|
| Claude Code | claude plugin marketplace add /path/to/whetstone-ai && claude plugin install whetstone@whetstone-ai, or ad hoc claude --plugin-dir ./plugin | /whetstone:learn … |
| Claude Desktop | download whetstone-<version>.plugin from a release (or build it with python3 package_release.py) and upload it in the plugin manager. Or ./install_skills.sh ~/.claude/skills | /learn … |
| DeepSeek Harness | ./install_skills.sh ~/.agents/skills, then restart npx @deepseek-ai/dsh web | /learn … |
| pi | pi install git:github.com/Vico304/whetstone-ai (or a local checkout: pi install /path/to/whetstone-ai, or ./install_skills.sh ~/.pi/agent/skills) | /skill:learn … |
| Codex | install plugin/ as a plugin | $learn … |
Type / and you should see guide / outline / map / learn / clarify. Everything the tool writes goes under whetstone/ inside your materials directory; the materials themselves are never modified.
| Skill | What it does |
|---|---|
| learn | Main entry. Fills in whatever is missing (your profile, a survey of the materials, a course-by-course plan, this course's outline), then generates one lesson document per section, checks prerequisites and tutors section by section |
| guide | Wizard. Asks whether you want a tour or a plan; planning records your background and goals, surveys a directory of materials, and confirms the course sequence one course at a time. Plans only |
| outline | Generate, discuss or revise one course's outline; writes it back after you confirm |
| map | Builds or revises a course's system map — parts, what contains what, labelled links — and turns a section into one that places the parts or follows one run; writes back on confirmation |
| clarify | Groups the [[wikilinks]] you leave in the lesson documents by unit and writes one source-grounded note per unit: a single unclear concept is just explained, several in one unit get the unit re-taught from them (Obsidian-compatible) |
Two choices are made while planning: full or fast (fast narrows the scope and relaxes judging, but still predicts before revealing); by material or by domain (by domain starts with a skeleton course of basic principles, with your materials as examples, and you pick branches to go deeper afterwards).
1.0, in daily use. The knowledge store is optional and has no long-term usage data yet.
cd plugin && python3 -m unittest discover -s tests
Standard-library Python only, no third-party dependencies.