convez.ai
SkillAutomaçãoIntermediário
lynxlangya

Techne — Skills Forçadas para Agentes IA

lynxlangya/techne

Skills que forçam investigação, julgamento e qualidade em agentes IA

Ver no GitHub
BomPluginScore · de 0 a 100

Biblioteca curada de skills e assets para agentes IA que força procedimentos cognitivos validados, não apenas prompts. Inclui skills como viz (diagramação de codebase), repro (reprodução de bugs), vet (code review estruturado) e intake (validação de briefs de engenharia). Funciona com Claude Code, Cursor, Gemini e outros hosts de agent skills.

Instalação

$npx convez-ai lynxlangya/techne

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
106
Forks
18
Acompanhando
9
Saúde
42%
Último commit
há 4 meses
Versão
—
Issues abertas
2
Contribuidores
1

Casos de uso

  • Investigar e diagramar arquitetura de codebases complexas
  • Forçar reprodução e verificação antes de corrigir bugs
  • Revisar código com scope computado e blast radius verificado
  • Validar briefs de engenharia antes de iniciar desenvolvimento

Veredito

Techne é um framework sólido que vai além de prompts genéricos, impondo workflows estruturados que reduzem erros em agentes IA. Recomendado para equipes que buscam qualidade verificável em code review, debugging e planejamento técnico.

README

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

techne

τέχνη · /ˈtɛk.niː/ · "TEK-nee"

Ancient Greek for craft · skill · art: the practical know-how of making and doing, and the shared root of both technique and technology. Aristotle placed technē (knowing how) beside epistēmē (knowing that).

Skills and agents for the AI era.

Website: techne.wangyun.fan.

Chinese version: README-CN.md.

What This Is

techne is a curated library of skills and agent assets for practical AI-era work. A techne skill is not a prompt snippet. It is a forced cognitive move: a small procedure that makes an AI collaborator investigate, judge, write, code, or ship in a way it often skips under pressure.

The project keeps one shared source of truth in skills/. Host-specific files stay thin: Claude Code plugin metadata, Cursor and Gemini manifests, and install instructions for Codex, Kimi, and other Agent Skills hosts all point back to the same skill bodies.

Skills

  • viz — investigate a real codebase by diagramming it. It routes one user request across architecture, interaction, data-model, state-model, and type-structure diagrams, then validates and builds a local viewer. See skills/viz/README.md for detailed usage.
  • repro — force bug fixes through a failing reproduction first, then verify with the same probe after the fix. See skills/repro/README.md for detailed usage.
  • vet — force code review through computed scope, accounted blast radius, verified claims/findings, and admissible verdicts before approving or requesting changes. See skills/vet/README.md for detailed usage.
  • intake — interrogate a written engineering implementation brief before work starts, accounting the fixed rubric, surfacing gaps/traps/questions, and emitting an intent-level plan. See skills/intake/README.md for detailed usage.

Install

Recommended: ask your AI agent. Paste this into Claude Code or Codex:

Install techne for my current agent by following https://github.com/lynxlangya/techne/blob/main/INSTALL.md, then verify the techne skills are available.

The agent reads INSTALL.md, detects your harness (Claude / Codex / Cursor / Gemini / Kimi), runs the right install path, and confirms the skill works.

Manual: per-harness commands and the universal npx skills fallback are in INSTALL.md.

For Claude Code:

/plugin marketplace add lynxlangya/techne
/plugin install techne@techne-dev
/reload-plugins

For Codex:

npx skills add lynxlangya/techne -a codex -g -y

See WORKFLOW.md for the delivery process and ROADMAP.md for the product map.

Documentation

Public-facing documentation is split by language:

  • README.md is the default English document.
  • README-CN.md is the Chinese companion document.

The same convention applies to skill usage guides when a Chinese version is needed, for example skills/viz/README.md and skills/viz/README-CN.md.

Ver o original no GitHub

Bom para

Relacionados