Teach any AI agent to think lean.
Agent skills for reducing cost, carbon, energy, and complexity — across your web, cloud, data, and AI stack. The same wasted compute drives up energy, cost, and carbon — three symptoms of one problem. These free, open-source skills teach your agent to find that waste, recommend fixes, and produce evidence-based reports you can trust.
Wasted compute has three symptoms.
The same oversized instance, bloated prompt, or uncached inference burns extra energy, inflates your cloud bill, and grows your carbon footprint — in the same act. These skills give your AI agent real expertise from the Green Software Foundation, so every audit addresses all three at once.
An expert for every layer
Oversized servers, heavy web pages, chatty databases, AI agents stuck in loops — your agent gains a specialist for each, trained on what to look for and how to fix it. A smart router brings in the right one for your question.
Answers you can trust
Every finding shows the evidence behind it. Real counts get reported; estimates are clearly labeled as estimates. If something can't be measured yet, the report says so — and tells you how to measure it.
Efficiency pays three ways
Energy is the physics; cost is how the work gets funded; carbon is why it matters. Every finding names the countable driver (tokens, gigabytes, instance-hours) so finance sees the payback, sustainability sees the impact, and engineering sees the underlying inefficiency — from one audit.
Ask a question. Get an honest answer.
You ask in plain language
"Is my website green?" "Why is our cloud bill growing?" "Make our chatbot cheaper." No commands to learn — the router understands the question and picks the right skill.
The right expertise kicks in
It checks your site, config, logs, or bills against proven efficiency patterns from the Green Software Foundation — and writes down what it found, with evidence.
Results become anything
All skills share one findings format, so any audit can become a slide deck, a written report, a live dashboard, sprint tickets, or a two-line summary for your release notes.
Nothing is ever made up. If it can be counted — tokens, megabytes, idle servers — it's counted and shown. If it can't be measured yet, the report says "not measured" and points you to the free tools that measure it for real. No fake percentages, no vague "green scores." That's why you can put these reports in front of your boss.
Three simple ideas turn skills into expertise.
No framework, no service, no database. Just a smart way of organizing expertise so the pieces snap together.
1 · The router
You never memorize skill names. One skill — lean-router — reads your question, picks the right specialist, and chains skills together when a job needs several. It even saves energy itself: easy steps get quick thinking, hard steps get deep thinking, and finished audits are reused instead of re-run.
2 · One skill, one job
Each skill is deliberately small: the web auditor only audits web pages, the deck builder only builds decks. Small skills are easy to trust, easy to test, and easy to improve. Every skill even ends by naming its neighbors — "not my job, ask region-selector" — so nothing overlaps.
3 · The shared contract
Every audit writes its results in the same simple file: lean-findings.json. That one decision is the magic — any audit can feed any report, deck, dashboard, or ticket export. Add a new audit skill, and every output skill works with it instantly.
{
"title": "Hero images not compressed",
"severity": "high",
"evidence": "4 images, 6.2 MB, no modern format",
"impact": "roughly halves page transfer", ← energy & carbon: directional
"cost_signal": {
"driver": "GB egress per month", ← what's billed
"observed": "4.2 GB from this page", ← counted, not invented
"direction": "cuts the driver ~half"
},
"fix": "convert to WebP/AVIF with srcset",
"effort": "low"
}
Why one format changes everything
Because every skill speaks it, results flow like Lego bricks: audit → deck, two audits → one report, this month vs last month → progress dashboard. Each finding carries the split that keeps every claim safe: energy and carbon impact stay directional, cost driver stays counted (tokens, GB, instance-hours). No skill fabricates dollars; every skill helps you find them.
And a skill? It's just a folder.
No code required. A skill is a folder with one Markdown file describing when to activate, what to check, and how to stay honest. If you can write a checklist, you can write a skill:
name: my-new-skill description: Use this when the user asks about… ## What to check - Signature 1, with the evidence to look for - Signature 2 … ## Stay honest - Report only what you observed - Say what's out of scope ## Not this skill's job - X belongs to neighbor-skill
The project even ships a skill that reviews new skills against these standards — quality control, automated and open.
The expertise, skill by skill.
49 skills, each a small folder your agent reads. Install only what your stack needs — browse by what you want your agent to be able to do:
Free, open, and yours to extend.
Give your agent this expertise in minutes: download the skills, drop them in, ask your first question. Same audit, three payoffs — energy, cost, carbon. MIT licensed.
Have expertise the catalog is missing? Contributing it is one folder that follows the shared format — and the project even includes a skill that reviews new skills against its own standards. Your knowledge becomes something every agent can use.
We hold this website to the same standard as the reports. Your browser measured these numbers just now: