Agent Reach Tech v1.1

CLI standalone para coding agents pesquisarem tecnologia, desenvolvimento, open source e cibersegurança. Substitui o Agent Reach upstream para o escopo tech.

Pacote completo: CLI + skill auto-trigger + servidor MCP (9 tools).

Instalação (< 10 min)

git clone git@github.com:ggampp/agent-reach-tech.git
cd agent-reach-tech
pip install -e ".[mcp]"
agent-reach-tech install --env=auto
agent-reach-tech doctor

Ou via pip (sem clonar):

pip install "git+https://github.com/ggampp/agent-reach-tech.git#egg=agent-reach-tech[mcp]"

Windows helper:

.\scripts\install.ps1

Matriz de canais

Canal	Backend	Obrigatório	Zero-config
github	gh CLI + REST API	Sim	Público sim
web	Jina Reader	Sim	Sim
youtube	yt-dlp	Sim	Sim
rss	feedparser	Sim	Sim
hackernews	Algolia API	Sim	Sim
lobsters	RSS / JSON	Sim	Sim
cve_nvd	NVD API 2.0	Sim	Sim*
osv	OSV.dev	Sim	Sim
exa_search	Exa / fallback	Opcional	EXA_API_KEY
reddit	pullpush.io	Opcional	Sim

* NVD sem key funciona; NVD_API_KEY melhora rate limit.

Comandos

# v1.1 — routing e pesquisa orquestrada
agent-reach-tech route "vale a pena usar engram?" --json
agent-reach-tech research oss engram --repo Gentleman-Programming/engram --report
agent-reach-tech research cve CVE-2024-1234 --report
agent-reach-tech research trends --report
agent-reach-tech mcp

# Canais
agent-reach-tech doctor
agent-reach-tech github repo OWNER/REPO
agent-reach-tech web URL
agent-reach-tech youtube info URL
agent-reach-tech rss list
agent-reach-tech rss read URL
agent-reach-tech search --profile evaluate_library --name engram
agent-reach-tech catalog search engram
agent-reach-tech reddit search "cve" --subreddit netsec

Ver docs/MCP-SETUP.md para configurar o servidor MCP no Cursor.

vs Agent Reach upstream

	Agent Reach Tech	Upstream
Foco	Tech/OSS/security	Geral + China social
Dependência	Nenhuma	pip upstream
Canais	10 curados	15+
Catálogo local	Sim	Não
Workflows	4 receitas	SKILL genérico

Opcional

$env:EXA_API_KEY = "..."   # busca semântica
$env:NVD_API_KEY = "..."    # CVE rate limit
gh auth login               # GitHub privado

Testes

pip install -e ".[dev,mcp]"
pytest                  # offline (55+ tests)
pytest -m network       # live APIs
.\scripts\benchmark.ps1

Licença

MIT