convez.ai
SkillEscrita & ConteúdoAvançado
gyasis

Chiron — Gerador Universal de Aulas com IA

gyasis/chiron

Gerador LLM universal de aulas interativas para aprendizado solo

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill Claude Code que gera aulas interativas auto-contidas em HTML a partir de qualquer fonte (repositório, PDF, CSV, paper). Suporta especialização em três domínios: código, medicina e linguagem. Utiliza arquitetura Q8 onde o agente Claude Code pai executa todas as etapas de LLM, com validação por Zod schemas e orquestração de tentativas.

Instalação

$npx convez-ai gyasis/chiron

1 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

Estrelas
2
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 2 dias
Versão
v0.1.0
Issues abertas
17
Contribuidores
1

Casos de uso

  • Criar aulas interativas a partir de repositórios de código, PDFs ou papers
  • Gerar conteúdo pedagogicamente especializado em code, medicina ou linguagem
  • Produzir lições HTML self-contained com quiz, spaced repetition e logging de fraquezas

Veredito

Chiron é um projeto ambicioso e bem-arquitetado para geração de aulas multi-domínio com IA, usando padrões robustos (pipeline 5-stage, validação Zod, SQLite per-lesson). Requer familiaridade com Claude Code e design pedagógico; documentação detalhada suporta extensão e customização por domínio.

README

A tradução para o português deste README ainda está na fila.

Chiron — Universal Lesson Generator

Branch Tasks

"A centaur. A teacher. The tutor of Achilles, Asclepius, Jason, and Heracles."

TL;DR

Chiron is a universal LLM-powered lesson generator for solo learners across code, medicine, and language. Drop in any source — repo, PDF, vocab CSV, research paper — and get a self-contained interactive lesson HTML.

Fresh clone? Read SETUP.md first. It covers prerequisites, install, build, and verification — including how a brand-new Claude Code session bootstraps the skill.

Foundational facts

  • Three co-equal domains: code, medicine, and language. Not a code-tutorial tool with the others bolted on. Specialized paths per domain are encouraged where pedagogy demands it.
  • Single learner = Gyasi. No multi-user features, no auth. SQLite per-lesson at <lesson-output-dir>/.chiron-state.db.
  • AI multi-persona content (peer learners + tutor + native speaker) inside one solo-learner's lesson — content-layer feature, NOT a multi-user system.

Documentation hierarchy

DocPurpose
skill/README.mdSkill how-to — quick-start, supported domains, drop-in extension
skill/SKILL.mdSkill descriptor — top-level skill manifest
prd/chiron_design_v1_2026-04-28.mdComprehensive design PRD (post-debate)
prd/universal_lesson_generator_2026-04-28.mdOriginal design PRD
specs/001-chiron-v1/Spec + design artifacts: spec.md, plan.md, data-model.md, tasks.md, contracts/
memory-bank/projectbrief.md, systemPatterns.md, techContext.md, activeContext.md
CLAUDE.mdProject directives — Claude Code agent guardrails

Architecture summary

Pipeline. Five-stage Ingest → Generate → Validate → Build → Assemble flow. Each stage has typed inputs/outputs, validators, and retry semantics. See specs/001-chiron-v1/contracts/pipeline-stages.md.

LLM architecture. Per Q8 of the design PRD, the parent Claude Code agent runs all text-LLM steps in its own context — Chiron's skill structures prompts, validates outputs against Zod schemas, and orchestrates retries. There is no separate text-LLM gateway process. Vision (diagram OCR, figure description) routes through mcp__gemini-mcp__interpret_image.

Output. A single self-contained lesson.html per lesson — vendored libs (MathJax + mhchem, Mermaid, optional Pyodide, ChalkAI for reactive-math, an abstract MoleculeRenderer) are inlined per FR-037, so opening the HTML in a browser works with no build step. Runtime state — quiz attempts, mastery, SR scheduling, weakness log, bookmarks — lives in .chiron-state.db (SQLite, per-lesson, single learner).

Heritage repos

Audited locally in ~/dev/audits/:

  • ~/dev/audits/codebase-to-course/ — HTML rendering shell (reused ~85% as-is)
  • ~/dev/audits/classbuild/ — 5-stage pipeline + typed-schema pedagogy
  • ~/dev/audits/ai-course-generator/ — concept DAG validator + LLM-as-advisor

Per-domain assessment formats

DomainPrimarySecondarySR card type
CodeMCQ + true/false + spot-the-bugdrag-and-drop matchingConcept flashcards
MedicineMCQ paired with USMLE/AMBOSS-style clinical vignetteagreement-matrix (always/sometimes/never)Disease/drug/mechanism cards
Language (DE/IT)Fill-blank + matchingsentence reordering, dictationVocabulary + sentence cloze cards

Mode A vs Mode B

ModeTriggerOutput
A — Course-style"turn this textbook chapter into a lesson"Multi-chapter scroll-snap HTML site, Coursera-style
B — Case-study-style"make this incident a teaching moment"3-act lecture (Evidence → 2 Lectures → Synthesis) — see ~/.claude/skills/case-study.md

Open PRDs affecting Chiron

  • ~/dev/prd/scratch/chiron_tts_provider_selection_2026-04-29.md — TTS provider selection (Gemini API primary, ElevenLabs fallback) for the Italian native-speaker persona. Lives in the global PRD librarian, not the project tree.

License

TBD — likely MIT. No LICENSE file committed yet.

Status / next steps

WIP — see specs/001-chiron-v1/tasks.md for current progress. Wave 12 in flight.

Why "Chiron"?

Chiron was unique among centaurs: civilized, learned, and the prototype tutor. He taught medicine to Asclepius, strategy and music to Achilles, navigation to Jason, and astronomy to Heracles. One teacher, many subjects, deep mastery.

Ver o original no GitHub

Bom para

Relacionados