dont-reinvent

Stop your AI coding agent from reinventing the wheel. Save the tokens instead.

A Claude Skill that checks Free vs. Build vs. Buy before writing non-trivial code, and actually vets what it finds (license, maintenance, security) instead of stopping at star count.

The package facts shown — express-rate-limit, MIT, 3.3k stars, 2 runtime dependencies — were verified against the repository. The token and duration figures are illustrative of the asymmetry, not measured benchmarks. Source composition in assets/demo-source/.

Install

npx skills add Emanuelel/dont-reinvent

Works with Claude Code, Cursor, Codex, Copilot and ~70 other agents. Add -g to install globally instead of into the current project.

Also listed on skills.sh.

Other ways in — by hand, or on claude.ai

Copy it into your Claude Code skills folder yourself:

git clone https://github.com/Emanuelel/dont-reinvent.git
cp -r dont-reinvent/skills/dont-reinvent ~/.claude/skills/

Claude Code picks it up within the session, no restart needed. To scope it to one project, copy the folder into that repo's .claude/skills/ and commit it.

claude.ai, Cowork, and cloud sessions don't read ~/.claude/skills/ on your machine; they load the skills enabled for your Claude account, and custom skills are uploaded as a zip. Download dist/dont-reinvent.zip, then go to Customize → Skills, click + Create skill, and upload it. Requires a plan with code execution enabled.

Why this matters more now, not less

Build vs. buy used to be a slow decision. A developer's time and salary were the real cost of "build," so people naturally checked npm, Stack Overflow, or a vendor page first. Buying or reusing was often the faster path, and that friction kept the decision honest.

Agentic coding changes the economics, but not in the direction it looks like at first. Now that an AI agent can generate a working implementation of almost anything on request, building feels free and instant, so the path of least resistance quietly became "just build it," every time, by default. It isn't actually free. Every non-trivial feature costs real tokens: not just the first draft, but the edge cases, the debugging rounds, the security holes nobody thought to check. The old instinct to check "has someone already solved this" used to happen almost automatically. Now it gets skipped entirely, because asking an agent to build feels like the path of least resistance.

The result: left to its own defaults, an AI coding agent will happily hand-roll a rate limiter, a multi-model LLM router, a site audit tool, or a social-automation feature completely from scratch, even when a free, mature, actively maintained open-source project already does the job, more safely, for a small fraction of the token cost.

This skill exists to interrupt that default, one decision at a time.

Free, Build, or Buy: three options, not two

Not "build vs. buy." That framing hides the option that matters most.

• Free: an existing open-source solution, actually vetted (license, maintenance, security), not just found
• Build: write it from scratch, estimated in token-effort tiers (Low / Moderate / High / Very high), never a fake precise number
• Buy: pay for a hosted API/service, priced in real dollars/month