Research-First

A Claude Code plugin / agent skill that enforces a research-first workflow: always search for existing tools, libraries, MCPs, templates, frameworks, datasets, and open-source resources before creating anything from scratch.

Built for Claude Code. Also compatible with Codex, Gemini CLI, Cursor, and any agent that reads SKILL.md.

Works for any workflow: development, writing, data analysis, ops, and research.

What it does

When you ask your agent to build, write, analyze, or automate something, this skill makes it:

1. Detect your existing context - reads config files, style guides, data sources, and current stack.
2. Search for existing solutions across multiple layers:
   • Connected MCPs, plugins, and skills
   • Similar open-source projects you can use, self-host, fork, or learn from (awesome-selfhosted, openalternative.co, alternativeto.net, GitHub topics and trending)
   • Official registries (Anthropic MCP registry, Smithery.ai, mcp.run, Glama.ai)
   • GitHub repos, awesome-* lists, and open-source tools
   • Templates for writing, data, and process tasks (Notion, Kaggle, Hugging Face, etc.)
   • Developer forums (Reddit, HN, GitHub Discussions) for unfiltered opinions
   • Free-tier services via free-for.dev
3. Validate rigorously - activity, adoption, security (CVEs), license, and fit.
4. Detect self-promotional bias - cross-references 3+ sources, flags "Top 10" blogs that rank their own product #1.
5. Recommend one clear solution with alternatives and trade-offs.
6. Only then create - using found resources instead of reinventing the wheel.

Why

Creating from scratch when a battle-tested solution already exists is the most expensive mistake in any workflow. A 5-minute search can save hours of work and weeks of maintenance.

Install

Claude Code (recommended)

Add the marketplace, then install the plugin:

/plugin marketplace add niveku/research-first-skill
/plugin install research-first@niveku-plugins

Once installed, the skill triggers automatically. No manual invocation needed.

Other agents (Codex, Gemini CLI, Cursor, etc.)

Copy the skill folder into your agent's skills directory:

# Example: Claude Code manual install (per project)
cp -r skills/research-first .claude/skills/

# Example: Claude Code manual install (global)
cp -r skills/research-first ~/.claude/skills/

# Other agents: check your agent's skill directory convention
cp -r skills/research-first <your-agent-skills-dir>/

Usage

Once installed, the skill activates when you ask your agent to build, create, or implement something. No special commands needed.

Development:

• "Add authentication to my Next.js app"
• "I need web scraping in Python"
• "Set up CI/CD with free hosting for my React app"
• "Is there a library for X?"

Writing / content:

• "Write a product requirements document"
• "Create a weekly status report template"
• "Draft a go-to-market strategy"

Data / analysis:

• "Build a sales dashboard"
• "Analyze customer churn in our dataset"
• "Find a model for sentiment analysis"

Ops / process:

• "Automate my deployment pipeline"
• "Create an onboarding checklist for new engineers"
• "Set up infrastructure monitoring"

What's inside

research-first-skill/
├── .claude-plugin/
│   ├── plugin.json          # Plugin manifest
│   └── marketplace.json     # Marketplace entry (for /plugin install)
├── skills/
│   └── research-first/
│       ├── SKILL.md         # The skill instructions (loaded first)
│       └── references/
│           ├── dev.md       # Dev-specific search patterns
│           ├── writing.md   # Writing/content templates
│           ├── data.md      # Data, ML, analysis resources
│           └── ops.md       # CI/CD, infra, SOP resources
├── .github/workflows/
│   └── validate-plugins.yml # CI: validates plugin.json + SKILL.md
├── README.md
├── CHANGELOG.md
├── CONTRIBUTING.md
└── LICENSE

The skill uses progressive disclosure: SKILL.md loads first with the core workflow, and domain-specific details in references/*.md load only when relevant.

Contributing

See CONTRIBUTING.md.

License

MIT