Other people's tools, read properly.

What this is

None of this is mine and none of it is for sale. It all comes from claude-code-templates, an MIT-licensed library by Daniel Ávila, with individual pieces written by dozens of other people. It is free, you can download it yourself in about a minute, and this page exists because the hard part is not getting it. The hard part is knowing which fifteen of the eight hundred are worth your time.

So I read these. Not skimmed. The notes below include the ones I would not bother with, because a list where everything is recommended is not a list, it is a menu.

Worth your time

avoid-ai-writing

Audits writing for the patterns that make text read as machine-made, and rewrites them. The biggest file in the whole library at 90KB, and the depth is real rather than padding. MIT, by Conor Bronsdon.

What sold me is that it argues against itself in the opening paragraphs. It points out that independent audits have found AI-detection false-positive rates above 60% on people writing in a second language, and tells you not to make it the sole basis for a decision that matters. A tool that warns you about its own failure mode is a tool built by somebody who has watched it fail.

Where it stops: it is an editor, not a writer. It will clean up a draft and it will not tell you the draft was not worth writing.

free-tool-strategy

A framework for planning a free tool that earns attention rather than just costing you a weekend. Covers what to build, how deep to go, and where the lead capture belongs.

This one is uncomfortably close to my own situation, which is how I know it is asking the right questions. It made me admit the generator has no email capture and that this is a choice I have not actually made, only postponed.

Where it stops: it assumes you are building the tool to feed a paid product that already exists. If you have not got one, it will happily help you plan the marketing for something imaginary.

schema-markup

Structured data. The invisible labels that tell a search engine a page is a product, or an article, or a person, rather than making it guess from the words.

I am recommending this partly because this site has none, which I found out while reading it. That is the useful kind of tool: one that shows you a gap rather than confirming what you already knew.

Where it stops: structured data helps a search engine understand a page. It does not make the page worth understanding. If nobody links to you, correct schema changes nothing.

accessibility-auditor

WCAG 2.1 in a usable shape: the four principles, the common failures, ARIA patterns, and a testing checklist that includes actually using a screen reader.

Solid, and broader than the contrast checking most people mean when they say accessibility. The keyboard and focus sections are the parts worth reading twice, because those are the failures you cannot see in a screenshot.

Where it stops: it audits against a standard. Passing WCAG and being pleasant to use with a screen reader are different achievements, and only the first one can be checked from a list.

page-cro

Walks a marketing page and suggests what to change to get more people doing the one thing the page is for. Sensibly, it makes you name that one thing first.

Where it stops: every recommendation is a guess until you measure it, and the skill cannot measure anything. It is a list of hypotheses. Treat it as one.

skill-creator

How to write a good Claude skill. Apache 2.0, from Anthropic's own repository, so it is the closest thing to a specification that exists.

The one to read if you want to stop collecting other people's components and write your own, which is the correct destination for anybody serious about this.

Useful, with a catch

seo-fundamentals

Reference tables for E-E-A-T and Core Web Vitals with the target numbers. Genuinely handy to have to hand.

The catch: at under 3KB it is a cheat sheet rather than a skill. Nothing wrong with that, but do not expect it to do any work for you.

pricing-strategy

Thorough on tiers, value metrics, willingness-to-pay research and when to raise prices.

The catch: it says SaaS in the first line and it means it. Tiers, trials, recurring revenue. If you sell a file once, as I do, a good half of it is answering a question you are not asking.

diagrammer

Turns a description of a system into a clean SVG diagram.

The catch: it needs a separate program installed on your machine first. Fine if you knew that going in, annoying if you find out halfway through.

I would skip these

writing-rules

Not what the name suggests. It is not about writing style at all, it is about configuring a hook tool called Hookify. Good enough at that, but if you went looking for guidance on prose you will read three paragraphs before realising.

code-review-excellence

Under 1.5KB and marked as community-contributed. Sound advice, in the sense that "review for correctness, security and performance" is sound advice. There is no method in it you did not already have.

humanizer

Only because it overlaps heavily with avoid-ai-writing, which is five times the size and considerably more careful. Pick one, and pick that one.

docx, pdf-anthropic, pptx, xlsx

These are not open source. The library's own attribution file describes them as source-available reference examples, not for redistribution, and points out they are point-in-time snapshots that are not maintained. Equivalents ship with Claude already, so there is nothing to gain and a licence to break.

How to actually use one

Copy the single folder you want into .claude/skills/ in your project. Not all of them. A skill is instructions that change how an agent behaves, so installing eight hundred is not a library, it is noise, and the useful ones get harder to find rather than easier.

Read it first, for the same reason. Somewhere between a quarter and a third of the files in that library are shell, Python or JavaScript. Installing one is closer to running code than to reading a document.

Get the library See what I built