The Claude skills that 10x your workflow on a token budget - grill-me, superpowers, impeccable, data + design skills, mapped to all 17 tracks. by Phoebe Fu
▶︎ Open the live course →
以下摘录自该仓库的 README(片段) · 在 GitHub 读完整 README →
Free, runs in your browser. No install, no login.
> 📚 Part of **Learn with Phoebe** - free, hands-on courses on AI, data, and the craft around them. **Browse every course ↗**
learn-claude-skills-with-phoebe
The Claude skills that 10x your workflow - on a token budget. A free, interactive course by
**Phoebe Fu**.
Most people use Claude like a very good chat box. The people who get 10x out of it use
**skills**: reusable instruction packs Claude loads only when a task needs them. Skills make a
workflow drastically better *and* spend fewer tokens doing it, because a skill's detail stays on
disk until the moment it fires (progressive disclosure). This course teaches the highest-leverage
skill stack, organized by all 17 of Phoebe's course-tracks.
The arc (8 sessions + a reference matrix)
1. **The Core Stack & token economics** - what a skill is, progressive disclosure, the 8 skills every track inherits
2. **Build & ship** - TDD, systematic-debugging, git worktrees, cavecrew, deploy (prog / deng / sys / docs)
3. **Data & DS** - schema-to-insights, the data-skill factory, dataviz, profiling (data / ds)
4. **AI & agents** - prompt + API + caching, office docs, skill-creator, artifacts (ai / aiap)
5. **AI in your domain** - marketing-skills, content-multiplexer, design-partner, deep-research (aidm x12)
6. **Content, viz & comm** - sketch family, the impeccable taste gate, decks, humanizer (viz / comm / biz)
7. **Lead, ops & edge** - mentor-room, hooks + schedule, safe edge tracks (lead / gov / emrg / prod / move)
8. **Build your own skill** - skill-creator, a token-lean SKILL.md, the budget playbook (capstone)
Plus **The 17x10 matrix** - top skills for every one of the 17 tracks, on one browsable page.
The thesis
Every technique reduces to five token moves: **don't load it** (progressive disclosure) ·
**offload it** (subagents, Explore) · **compress it** (caveman, deep-research) · **get it right
once** (grilling, verification) · **reuse it** (Projects, prompt caching).
Sources
Built on Anthropic's Agent Skills material (DeepLearning.AI + Anthropic Academy), the prompt