Fable Teach Skills
az9713/fable-teach-skills
Workspaces de aprendizado estruturados gerados por /teach em múltiplos domínios
Ver no GitHubColeção de workspaces de ensino estruturados gerados pela skill /teach do Matt Pocock em Claude Code. Cada domínio (malabarismo, canto, poesia Tang) demonstra como uma única ferramenta /teach cria currículos missão-orientados e autocontidos em diferentes áreas de conhecimento.
Instalação
npx convez-ai az9713/fable-teach-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Aprender malabarismo, canto ou poesia Tang através de lições estruturadas
- Estudar arquitetura de currículos educacionais baseados em missões
- Stress-test de ferramentas de IA para ensino multidisciplinar
Veredito

Projeto engenhoso que demonstra como a skill /teach cria estruturas de aprendizado reutilizáveis e centradas em missões. Funciona melhor como referência de design educacional do que como skill pronta para produção.
README
A tradução para o português deste README ainda está na fila.
fable-teach-skills — a /teach stress-test collection
▶️ Open the landing page →
A growing collection of teaching workspaces produced by running Matt Pocock's
/teach skill inside Claude Code (model: Claude Fable 5).
Each subfolder is one independent, self-contained /teach run in a different domain — a
stress test of how well a single slash command turns into a structured, mission-grounded
curriculum across very different subjects.
Domains
| Domain | Request | Mission | Live lesson |
|---|---|---|---|
| 🤹 Juggling | /teach how to juggle 3 balls | Land a 3-ball cascade as a party trick | Lesson 1 → |
| 🎵 Singing | /teach how to sing like Luciano Pavarotti | Sing one iconic tenor aria, proudly | Lesson 1 → |
| 📜 Tang poetry | /teach me to write a "tang shi" like 唐詩三百首 | Compose a 律詩 worthy of the anthology | Lesson 1 → |
More domains will be added as further stress tests are run.
What /teach is
/teach treats the current directory as a stateful learning workspace that persists across
sessions, rather than answering a one-shot "explain X" prompt. Its philosophy: deep learning needs
three things —
- Knowledge — captured from high-trust sources, never the model's memory (
RESOURCES.md) - Skills — acquired through tight interactive feedback loops (
lessons/) - Wisdom — tested against real practitioners (communities in
RESOURCES.md)
Everything is anchored to a single mission — the real reason the learner cares.
Repository layout
.
├── index.html # Landing page (links into every domain)
├── README.md # You are here
├── <domain>/ # One self-contained /teach workspace per domain
│ ├── MISSION.md # The "why" — the compass for every teaching decision
│ ├── RESOURCES.md # Trusted knowledge sources + practitioner communities
│ ├── NOTES.md # Learner preferences + working notes
│ ├── index.html # Redirect to that domain's current lesson
│ ├── lessons/ # Self-contained, beautiful HTML lessons
│ ├── reference/ # Compressed cheat sheets & glossaries (revisited often)
│ └── learning-records/ # Evidence-based records of what was actually learned
Adding a new domain
- Create a new folder (e.g.
pottery/) and run/teach …with that folder as the working directory. - Add a
<domain>/index.htmlredirect to its first lesson. - Add one row to the Domains table above and one card to
index.html.
Development journal
Domain 3 — 唐詩 Tang regulated verse (📜 tang-shi/)
The prompt:
"Teach me how to write a 'tang shi' like those in Tang Shi San Bai Shou. After you teach me, I should be able to write a 'tang shi' that will still be celebrated after hundreds if not thousands of years."
This was the most demanding /teach stress-test in the collection so far, for two reasons:
(1) the subject domain — Tang regulated verse — has a phonological dimension (Middle Chinese tones)
that cannot be recovered from modern fluency alone; (2) the stated ambition — poetry celebrated for
a millennium — set up a mission-grounding moment: Fable 5 had to honestly bound what is and isn't
trainable before writing a single lesson.
How Fable 5 scoped the mission
After learning the learner reads Traditional Chinese fluently (including some classical) and
deliberately chose 律詩 (the hardest 8-line regulated form) over the simpler 絕句 quatrain, the
model wrote a MISSION.md that frames the aspiration honestly: canonization is decided by readers,
history, and luck — it cannot be trained. What can be trained is the craft that gave the 300 poems
their chance. Success is defined as passing 搜韻's 律詩校驗 with zero errors,
writing parallel couplets a literate reader credits as real craft, and submitting a finished poem to
a practitioner community for critique.
The zone of proximal development decision for lesson 0001
Rather than opening with poetic imagery or thematic content — the natural "interesting" entry point — Fable 5 identified 平仄 (the tonal classification of every syllable as either level or oblique) as the correct first rung, for a specific reason: modern Mandarin fluency actively misleads you here. The entire 入聲 (entering tone) category — short syllables historically ending in -p, -t, -k — dissolved into all four modern tones when Middle Chinese evolved into Mandarin. This means words like 白, 國, 別, 月, 不 sound level in Mandarin (bái, guó, bié, yuè, bù) but are oblique in Tang prosody. A learner who skips this step would mark half the tonal grid wrong without knowing why.
How lesson 0001 was constructed
Fable 5 verified its primary sources before building anything:
| Source | Role | Verified |
|---|---|---|
| 唐詩三百首 — UVa Etext | Anchor anthology (the learner's stated target) | ✓ |
| 搜韻 sou-yun.cn | 平水韻 tables + automated 律詩校驗 (the feedback loop for compositions) | ✓ |
| 全唐詩 — ctext.org | Full Tang corpus in 繁體 for wider exemplar lookup | ✓ |
| 漢典 zdic.net | Per-character phonology incl. 廣韻 fanqie + 平水韻 category | ✓ (blocks automated fetches; browser-verified reference) |
| 詩詞吾愛 52shici.com | Practitioner community with 格律檢測 tools | ✓ |
| 王力《詩詞格律》(中華書局) | Canonical modern manual — the workspace's authoritative ruling | Print/library |
| 韻典網 ytenx.org | 廣韻-level character lookup | ✗ — TLS cert expired; excluded and noted in RESOURCES.md Gaps |
Du Fu's 《春望》was chosen as the practice poem because it is a 五言律詩 — the exact form the learner is targeting — and because it is unusually dense with teaching material: seven 入聲 characters, a reverse trap (勝 read as 平 in 不勝簪), and a polyphone (更). The lesson is a click-to-mark interactive drill: the learner taps each of the 40 characters to label it ○ (level) or ● (oblique), then checks against the 平水韻 answer key. Wrong characters receive per-character annotation explaining why — including which ancient rime category the character belongs to and what the equivalent sounds like in topolects (Cantonese, Minnan) that still preserve the entering tone. A perfect score unlocks the rhyme analysis: 深心金簪 = 下平十二侵, illustrating the one-group rhyme rule for all future lessons.
Reference documents created alongside lesson 0001
reference/pingze-cheatsheet.html— Mandarin→平仄 decision rules, high-frequency 入聲 watchlist, polyphone table (勝更看思騎過), and links to the two tool-verified arbiters.reference/glossary.html— ~20 terms (四聲, 平仄, 入聲, 平水韻, 黏對, 孤平, 拗救, 對仗, 起承轉合, 用典, 詩眼…) with lesson-tagging; all future lessons adhere to this vocabulary.
Planned lesson arc (from tang-shi/NOTES.md, calibrated per session by learning-records):
平仄 ✓ → four line types & 黏對 → 押韻 / 平水韻 groups → first 五絕 as scaffolding exercise → 對仗 → composing a full 五律 → 拗救 → 用典 & imagery → 七律
Credits
- Teaching method:
/teachfrom mattpocock/skills by Matt Pocock. - Workspaces generated in Claude Code with Claude Fable 5.