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.
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.
| 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 |
cursor-code-documentation |
Cursor IDE | Native plugin system | plugins/cursor-code-documentation/README.md |
| Standalone | Any AI tool | npx skills add / manual |
skills/README.md |
- 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%.
- 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.
- A Complete Guide to AGENTS.md — Progressive disclosure patterns, monorepo support, domain files.
For a full evidence-to-implementation mapping, see DESIGN-GUIDELINES.md.
/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.
/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.
/plugin marketplace add rodrigorjsf/agent-engineering-toolkit
/plugin install orchestrate@agent-engineering-toolkit→ See plugins/orchestrate/README.md for prerequisites and the configuration reference.
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.
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.
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-code-documentation/README.md for full setup instructions.
npx skills add rodrigorjsf/agent-engineering-toolkit -g→ See skills/README.md for per-tool install options and manual installation.
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.
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.
MIT