Tutor de Álgebra 1 para Claude
RealRogerWinter/algebra-1-tutor
Tutor de álgebra com diagnóstico de misconceptions e verificação automática
Ver no GitHubUm tutor interativo e paciente de Álgebra 1 que se integra ao Claude como skill, oferecendo ensino personalizado, diagnóstico de erros conceituais e verificação de respostas com SymPy. Acompanha também um textbook gratuito online com exemplos resolvidos, gráficos computados e referência de código para cada tópico.
Instalação
npx convez-ai RealRogerWinter/algebra-1-tutor1 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 álgebra do zero para adultos com base em aritmética
- Preparação para testes e provas de álgebra
- Diagnóstico e correção de erros conceituais específicos em operações algébricas
- Prática autoguiada com feedback imediato e gráficos renderizados
Veredito

Uma implementação rigorosa de tutoria adaptativa que diagnostica misconceptions em vez de apenas marcar respostas como erradas. Funciona no plano gratuito do Claude e complementa bem quem precisa de reforço estruturado em álgebra, com a vantagem de código verificado via SymPy e fonte única de verdade entre tutor, textbook e prática.
README
A tradução para o português deste README ainda está na fila.
Algebra 1 Tutor
A complete, self-paced Algebra 1 course for an adult who knows arithmetic and is meeting the rest of algebra for the first time. It comes as a patient tutor you install into Claude (a one-on-one teacher that works problems with you, diagnoses why a wrong answer went wrong, checks every answer, and draws the graphs) and as a free browser textbook you can read with no account. Both are built from one machine-verified source.
Install the tutor in Claude
Download algebra-1-tutor.zip, then in the Claude desktop or mobile app or at claude.ai:
- Click Customize in the left menu, then open Skills.
- Click the + at the top, choose Create Skill → Upload a skill, and select the
.zip. - Start a chat and say what you want:
- "I'd like to learn algebra from the beginning."
- "Help me with slope and y = mx + b, I have a test Friday."
- "Walk me through 5.3.4." Every item in the textbook has a short code you can quote.
- Paste a Progress Card to resume a previous session.
It works on the free plan (today backed by Sonnet 4.6). Larger models such as Opus tutor a little more sharply but aren't required. Code execution is recommended but optional: with it the tutor runs sympy to check answers and compute graphs; without it the tutor falls back to careful written-out substitution. The skill triggers on algebra topics even when you don't say "algebra": "solve for x", "factor this", "what's a function".
Using Claude Code?
Install the same tutor from the terminal:
/plugin marketplace add RealRogerWinter/algebra-1-tutor
/plugin install algebra-1-tutor@algebra-1-tutor
The tutoring works the same, but graphs and $$…$$ notation don't render in a terminal: math shows as raw text and figures as raw SVG markup, so the app or website is the smoother place to learn. New to skills? The how-to-use guide walks through setup and a first session.
Prefer to just read?
The whole course is a free website (no account, no install): https://realrogerwinter.github.io/algebra-1-tutor/. Worked examples to study, practice with per-question answers you can reveal, computed graphs, and a short reference code beside each item so you can point the tutor at one exact thing. It also includes a student guide (how to learn with the course) and a tutor guide (extra parallel-form practice).
Every answer in the course is checked with sympy, and the whole thing is generated from a single source of truth, so the textbook, the tutor, and the practice sets always agree.
What it teaches
13 units and 52 lessons, from foundations to a quadratics capstone, plus a core unit on data and statistics. Full map in docs/CURRICULUM.md.
- Go in order or jump anywhere. Every unit declares its prerequisites, so jumping to "slope" is safe — the tutor patches gaps instead of forcing a restart.
- Diagnoses misconceptions. It reads the specific wrong answer (inverting rise over run, dropping a negative, reading
=as "compute") and repairs it with a targeted question and a fresh representation, rather than just marking it wrong. - Verifies before asserting. It checks every answer before calling it right or wrong, and assumes you may be the one who's right on a disagreement. No confidently mis-graded correct answers.
- Shows the math. Notation in
$$…$$LaTeX plus plain Unicode (x², √12, ½) inline, and graphs as computed, labeled artifacts with a companion table. - Remembers across sessions through a copy-paste Progress Card, since Claude chats are otherwise stateless.
How it teaches (design in brief)
- Concrete → pictorial → symbolic — a balance scale for equations, money and debt for negatives, an input-output machine for functions, then the symbols.
- Ask before telling, with a graduated hint ladder that still honors "just tell me."
- Backward-faded worked examples — show one fully, then hand back more of the work as you get fluent.
- Check understanding by transfer, not "does that make sense?"
- Encouragement decoupled from correctness — acknowledge effort plainly; never praise a wrong result.
Full rationale and file layout: docs/ARCHITECTURE.md.
How it's built and verified
The course is generated from one source of truth and held correct by a CI gate, so the four surfaces (textbook, tutor skill, student guide, tutor guide) can't silently drift apart.
curriculum.yamlis the structural source of truth (units, lessons, prerequisites).- Lesson prose lives in two synced copies: tutor-facing (
algebra-1-tutor/references/units/) and student-facing (textbook-src/). - Every computational answer is stored as machine-readable data and re-derived with
sympy:
python _verification/verify_answers.py
# Files checked: 13 | Total problems: 909 | auto-checked: 614 | manual (skipped): 295 | Failures: 0
python _verification/verify_complementary.py
# Complementary files: 13 | problems: 269 | auto-checked: 214 | manual: 55 | failures: 0
The "manual" problems are conceptual (classify a number, function-or-not, word-problem setups): answered in the lessons but not auto-checkable. Totals grow as content is added; the invariant that matters is 0 failures, enforced on every push by CircleCI along with alignment, notation, figure-accuracy, byte-identical-build, and packaging checks plus a pytest suite.
See docs/DEVELOPMENT.md to edit, re-verify, and repackage, and CONTRIBUTING.md for the full workflow.
Repository layout
.
├── .claude-plugin/marketplace.json # Claude Code plugin marketplace (enables /plugin install)
├── curriculum.yaml # Source of truth: units, lessons, prerequisites
├── algebra-1-tutor/ # The skill that gets packaged & uploaded (CC BY-NC)
│ ├── SKILL.md # Operating manual: persona, pedagogy, verification, notation
│ ├── references/ # curriculum-map, misconceptions, metaphors, visuals, pedagogy, sources
│ │ └── units/ # One file per unit (12 numbered + Unit A statistics = 13)
│ └── figures/ # 22 computed, sympy-checked SVG figures (bundled in the skill)
├── textbook-src/ # Student-facing prose for the textbook (same math as the tutor)
├── _verification/ # Build & verification tooling (MIT; not shipped in the skill)
│ ├── build_all.py # Regenerate the textbook, student guide, tutor guide, and landing
│ ├── build_skill.py # Package algebra-1-tutor/ → algebra-1-tutor.zip
│ ├── verify_answers.py # sympy re-check of every answer key
│ ├── verify_complementary.py # sympy re-check of the tutor-guide practice sets
│ ├── check_alignment.py # SSOT / notation / figure / reference-code guards
│ ├── figures.py # Computed, accuracy-checked SVG figure library
│ ├── unit-*.json # Machine-checkable answers; complementary/ holds the tutor sets
│ ├── AUTHOR_GUIDE.md # Unit-file authoring spec
│ └── tests/ # pytest suite
├── docs/ # Generated GitHub Pages site (CC BY-NC) + project docs
│ ├── index.html # Landing page
│ ├── textbook/ student-guide/ tutor-guide/ # the generated reader sites
│ ├── assets/ # Hero & illustration images (CC0; see assets/README.md)
│ └── ARCHITECTURE.md DEVELOPMENT.md CURRICULUM.md
├── algebra-1-tutor.zip # Built, installable skill package
├── LICENSE LICENSE-CONTENT # MIT (code) + CC BY-NC 4.0 (content)
└── README.md
Why the skill is built this way (platform constraints)
The design works around verified constraints of the Claude.ai / app surface:
- No native image generation, and raw SVG in a chat message doesn't render → graphs are emitted as Artifacts with computed coordinates.
- Only
$$…$$LaTeX renders dependably on Claude.ai; inline\(…\)and$…$often show as raw text → real notation goes in$$…$$blocks, inline math is plain Unicode. - Conversations are stateless across chats → the Progress Card carries progress.
- A 13-unit course can't live in one file → a lean
SKILL.mdplus per-unit reference files loaded on demand.
Contributing
Issues and pull requests are welcome — a wrong answer, a clearer explanation, a better figure. The math is machine-checked and the four surfaces are kept in sync by CI, so start with CONTRIBUTING.md, which walks the edit-and-verify loop.
License
Two licenses, by directory:
- Code and build tooling — MIT. Covers
_verification/,curriculum.yaml,requirements.txt,.circleci/. See LICENSE. - Course content — Creative Commons CC BY-NC 4.0 (free to share and adapt for non-commercial use, with attribution). Covers
algebra-1-tutor/,textbook-src/, and thedocs/site. See LICENSE-CONTENT. - AI-generated images in
docs/assets/are CC0. See docs/assets/README.md.
The course content is original; OpenStax and other sources are cited as further reading, not reproduced.
Status
Functional, deployed, and verified. The tutoring behavior was validated through simulated student sessions; the real proof is live use on Claude.ai. Feedback and fixes welcome.
Built with the skill-creator workflow.