Agent Engineering Toolkit
rodrigorjsf/agent-engineering-toolkit
Gera arquivos mínimos de config para agentes com base em evidências científicas
Ver no GitHubConjunto de plugins para gerar e otimizar arquivos de configuração AGENTS.md e CLAUDE.md para agentes de IA. Utiliza princípios de divulgação progressiva baseados em pesquisa acadêmica (ETH Zurich) para criar arquivos mínimos e focados, evitando configurações genéricas inchadas que reduzem performance.
Instalação
npx convez-ai rodrigorjsf/agent-engineering-toolkit1 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
- Otimizar contexto e instruções para agentes Claude Code
- Configurar agentes em Cursor IDE com máximo de eficiência
- Reduzir tamanho de arquivos de contexto mantendo qualidade
- Implementar progressive disclosure em projetos complexos
Veredito

Ferramenta robusta e fundamentada em pesquisa científica real (não marketing). Custo inicial alto em tokens, mas investimento único que melhora permanentemente futuros agentes. Ideal para times que constroem agentes IA produção com foco em efficiency.
README
A tradução para o português deste README ainda está na fila.
Agent Engineering Toolkit
A multi-plugin marketplace providing evidence-based agent artifact engineering. Instead of auto-generating one bloated configuration file, this toolkit creates minimal, scoped files following progressive disclosure principles — proven by research to outperform comprehensive auto-generated configurations. Ships as Claude Code and Cursor distributions — each an Initializer/Customizer pair — plus a standalone distribution compatible with any AI coding tool.
Cost and Model Guidance
These plugins analyze your entire codebase before generating or improving configuration files. Execution cost scales with project size and scope — a large or complex project can be expensive to run.
Recommended model: Claude Opus delivers the best analysis quality for this workload. Viable alternative: Claude Sonnet with High effort produces decent results at lower cost.
Usage pattern: run each skill once per project, or when the codebase has changed significantly. Not on every session.
The long-term benefit is that every future agent session becomes cheaper and more accurate after the generated files are in place. Treat the first execution as a one-time investment — not routine work.
Distributions
| Distribution | Platform | Install Path | Full Documentation |
|---|---|---|---|
agents-initializer | Claude Code | Native plugin system | plugins/agents-initializer/README.md |
agent-customizer | Claude Code | Native plugin system | plugins/agent-customizer/README.md |
orchestrate | Claude Code | Native plugin system | plugins/orchestrate/README.md |
cursor-initializer | Cursor IDE | Native plugin system | plugins/cursor-initializer/README.md |
cursor-customizer | Cursor IDE | Native plugin system | plugins/cursor-customizer/README.md |
| Standalone | Any AI tool | npx skills add / manual | skills/README.md |
Research Foundation
Academic Research
- Evaluating AGENTS study (ETH Zurich, Feb 2026) — The first rigorous study of context file effectiveness. LLM-generated files reduce performance by 3%; developer-written minimal files improve it by 4%.
Anthropic Official Documentation
- Context Engineering for AI Agents — Defines "context rot" and the attention budget concept.
- CLAUDE.md Memory Documentation — Recommends under 200 lines per file.
- Best Practices — "If Claude already does something correctly without the instruction, delete it."
- Lost in the Middle (TACL 2023) — Models perform worst on information buried in the middle of long contexts.
Practitioner Guides
- A Complete Guide to AGENTS.md — Progressive disclosure patterns, monorepo support, domain files.
For a full evidence-to-implementation mapping, see DESIGN-GUIDELINES.md.
Installation
Claude Code — agents-initializer
/plugin marketplace add rodrigorjsf/agent-engineering-toolkit
/plugin install agents-initializer@agent-engineering-toolkit
→ See plugins/agents-initializer/README.md for scope flags and full options.
Claude Code — agent-customizer
/plugin marketplace add rodrigorjsf/agent-engineering-toolkit
/plugin install agent-customizer@agent-engineering-toolkit
→ See plugins/agent-customizer/README.md for scope flags and full options.
Claude Code — orchestrate
/plugin marketplace add rodrigorjsf/agent-engineering-toolkit
/plugin install orchestrate@agent-engineering-toolkit
→ See plugins/orchestrate/README.md for prerequisites and the configuration reference.
Cursor IDE — cursor-initializer
git clone https://github.com/rodrigorjsf/agent-engineering-toolkit.git ~/src/agent-engineering-toolkit
mkdir -p ~/.cursor/plugins/local
ln -s ~/src/agent-engineering-toolkit ~/.cursor/plugins/local/agent-engineering-toolkit
→ See plugins/cursor-initializer/README.md for full setup instructions.
Cursor IDE — cursor-customizer
git clone https://github.com/rodrigorjsf/agent-engineering-toolkit.git ~/src/agent-engineering-toolkit
mkdir -p ~/.cursor/plugins/local
ln -s ~/src/agent-engineering-toolkit ~/.cursor/plugins/local/agent-engineering-toolkit
→ See plugins/cursor-customizer/README.md for full setup instructions.
npx skills add — Standalone
npx skills add rodrigorjsf/agent-engineering-toolkit -g
→ See skills/README.md for per-tool install options and manual installation.
Repository Structure
agent-engineering-toolkit/
├── .claude-plugin/
│ └── marketplace.json # Marketplace catalog (Claude Code plugin system)
├── .cursor-plugin/
│ └── marketplace.json # Marketplace catalog (Cursor plugin system)
├── plugins/
│ ├── agents-initializer/ # Claude Code plugin — AGENTS.md and CLAUDE.md init/improve
│ │ ├── .claude-plugin/plugin.json
│ │ ├── README.md # Full plugin documentation
│ │ ├── skills/ # 4 skills: init-agents, init-claude, improve-agents, improve-claude
│ │ └── agents/ # 3 subagents: codebase-analyzer, scope-detector, file-evaluator
│ ├── cursor-initializer/ # Cursor IDE plugin — rules-first .cursor/rules/*.mdc init/improve (legacy AGENTS.md migration only)
│ │ ├── .cursor-plugin/plugin.json
│ │ ├── README.md # Full plugin documentation
│ │ ├── skills/ # 2 skills: init-cursor, improve-cursor
│ │ └── agents/ # 3 subagents (Cursor-native format)
│ ├── cursor-customizer/ # Cursor IDE plugin — single-artifact CRUD (rules, hooks, skills, subagents)
│ │ ├── .cursor-plugin/plugin.json
│ │ ├── README.md # Full plugin documentation
│ │ ├── docs-drift-manifest.md # Registry: reference files → source docs
│ │ ├── agents/ # artifact-analyzer + per-type evaluators (Cursor-native format)
│ │ └── skills/ # 8 skills: create-{type} and improve-{type}
│ ├── agent-customizer/ # Claude Code plugin — artifact creation and improvement
│ │ ├── .claude-plugin/plugin.json
│ │ ├── README.md # Full plugin documentation
│ │ ├── docs-drift-manifest.md # Registry: reference files → 12 source docs
│ │ ├── agents/ # 6 subagents: artifact-analyzer, evaluators, drift-checker
│ │ └── skills/ # 8 skills: create-{type} and improve-{type}
│ └── orchestrate/ # Claude Code plugin — autonomous backlog orchestration
│ ├── .claude-plugin/plugin.json
│ ├── .mcp.json # Registers the orchestrate-mcp server
│ ├── README.md # Full plugin documentation
│ ├── hooks/ # context-watchdog PostToolUse hook
│ ├── skills/ # 1 skill: orchestrate
│ ├── agents/ # 8 subagents: {investigator,implementer,reviewer,conflict-resolver}-{standard,deep}
│ ├── templates/ # commands.json, routing.json, handoff.json
│ └── orchestrate-mcp/ # MCP server (worktree, capability, routing, render, search tools)
├── skills/ # Standalone distribution — npx skills add compatible
│ ├── README.md # Full standalone documentation
│ └── {init,improve}-{agents,claude,skill,hook,rule,subagent}/
├── docs/ # Research and reference corpus
│ ├── claude-code/ # Claude Code specific docs
│ ├── cursor/ # Cursor IDE specific docs
│ ├── general-llm/ # Cross-tool research and guides
│ ├── shared/ # Cross-tool standards (Agent Skills)
│ └── analysis/ # Deep extraction analysis
├── DESIGN-GUIDELINES.md # Evidence-to-implementation mapping
└── LICENSE
Two separate skill sets by design:
plugins/*/skills/— plugin skills delegate analysis to isolated subagents, keeping the orchestrating context clean. Require the platform's plugin system.skills/— standalone skills perform all analysis inline. Compatible with any AI coding tool.
Contributing
Development conventions are enforced by .claude/rules/ — path-scoped rules load automatically when editing matching files:
plugin-skills.md— plugin skill authoring constraints (delegation, validation, limits)cursor-plugin-skills.md— Cursor plugin constraints (.mdc format, readonly agents)standalone-skills.md— standalone skill constraints (inline analysis, distribution awareness)agent-files.md— subagent file requirements (frontmatter, model, tools)readme-files.md— README structure and cost-warning requirements
See DESIGN-GUIDELINES.md for the evidence base behind each convention.
License
MIT