Canary — Agente de Automação de Testes com IA
bop-clocktower/canary
Gere e revise testes em múltiplos frameworks via linguagem natural
Ver no GitHubAgente Python que transforma requisitos em linguagem natural em testes de alta qualidade, automaticamente classificando o tipo (E2E, API, unit, performance) e recomendando o framework ideal (Playwright, Vitest, Pytest, k6). Funciona como plugin Claude Code sem API keys adicionais, usando a sessão de autenticação nativa.
Instalação
npx convez-ai bop-clocktower/canary1 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
- Gerar testes E2E e API a partir de descrições em português
- Revisar e desflakear testes existentes em Playwright/Vitest/Pytest/k6
- Recomendar o framework mais adequado para cada tipo de teste
- Scaffolding e migração de suites de teste para harness padronizado
Veredito

Solução madura e prática que reduz atrito na escrita de testes, integrando-se nativamente ao Claude Code sem dependências externas. Cobertura sólida dos principais frameworks modernos, embora a qualidade dos testes ainda dependa da clareza dos requisitos iniciais.
README
A tradução para o português deste README ainda está na fila.
Canary
Canary is an AI-powered test automation agent that transforms natural language requirements into high-quality, framework-aware test code.
🚀 Features
- Intent Classification: Automatically detects if you need E2E, API, Unit, or Performance tests.
- Framework Recommendation: Suggests the best-in-class tool (Playwright, Vitest, Pytest, k6) based on your needs.
- Code Generation: Produces production-ready test scripts following industry best practices.
- CLI-First: Simple terminal interface for seamless developer workflow.
🛠 Installation
pipx (recommended)
pipx install git+https://github.com/bop-clocktower/canary@latest
This installs the canary binary globally without polluting your project
virtualenv. Upgrade by re-running with @<new-tag>.
From source
git clone https://github.com/bop-clocktower/canary.git
cd canary
pip install -e .
Claude Code plugin
Canary ships as a Claude Code plugin with four MVP personas — canary-test-author,
canary-test-reviewer, canary-framework-advisor, and canary-flake-hunter —
plus the harness MCP server. Add it as a local marketplace:
/plugin marketplace add https://github.com/bop-clocktower/canary
/plugin install canary@bop-clocktower
Or, if you've already cloned the repo:
/plugin marketplace add /path/to/canary
/plugin install canary@bop-clocktower
Note: the install syntax is
<plugin-name>@<marketplace-name>. Thecanaryplugin ships from thebop-clocktowermarketplace, so the@qualifier reads ascanary@bop-clocktower.
No separate API key is required. The plugin runs through Claude Code's own session authentication.
📖 Usage
Generate Tests (Claude Code plugin)
Test generation runs inside Claude Code — no API key, no LLM provider setup. Invoke a persona from your editor session:
/canary-write-test Create a Playwright test for the login flow on https://example.com
The canary-test-author agent classifies the intent, picks a framework, reads
your repo for context via the Canary MCP tools, and writes the test using your
Claude Code session.
Recommend a Framework (CLI)
The CLI is deterministic and needs no key. To see Canary's framework recommendation and reasoning for a requirement:
canary recommend "I need a load test for my checkout API"
Other deterministic CLI commands: canary init <framework> (scaffold a suite),
canary run <file> <framework> (execute a test), canary migrate (adopt a
harness-scaffolded project), canary version.
⚙️ Configuration
No API key is required. Canary runs as a Claude Code plugin and uses your
Claude Code session for any LLM work — there is no separate provider key to
set. The CLI commands (recommend, init, run, migrate) are deterministic
and make no LLM calls.
📝 Changelog
See CHANGELOG.md for release history. Per-release notes are also published on GitHub Releases.
🏗 Architecture
Canary follows a modular pipeline: User Prompt → Classifier → Recommender → host LLM (your Claude Code session) → Generated Test
Generated by Canary — The future of autonomous test engineering.