Why findskills.co focuses on source-verified skills

There are two ways to build a skill directory.

You can publish a lot of entries quickly. Names, descriptions, install commands — all plausible-looking. Or you can publish only what you can trace back to a real upstream source.

findskills.co chose the second path. Here’s why it matters.

Most AI tool directories have a sourcing problem

The entries look real. They have names, descriptions, categories. But dig one level deeper and many of them are either generated, copied without attribution, or pointing to repositories that don’t exist anymore.

This matters more for skills than for other tools. A skill sits inside your agent’s context window and shapes how it reasons. Installing something from an unknown or fabricated source isn’t harmless. You’re changing how your agent approaches work — and you have no idea what process you actually installed.

The failure mode is subtle. A poorly-sourced skill doesn’t usually produce obviously wrong output. It produces plausibly structured output that follows a process you never reviewed and can’t trace back to any real-world expertise. That’s harder to catch than an obvious error.

What “source-verified” means here

Every listed skill points to:

• A real GitHub folder with a real commit history
• A raw SKILL.md file you can read before installing
• A visible upstream path (e.g. engineering-team/senior-architect/SKILL.md)
• A visible source repo and license

Fewer total listings than a speculative directory. Much stronger trust per listing.

Skills that couldn’t be verified were removed rather than kept as plausible-looking entries. The directory is smaller because of it. That’s a trade we’re willing to make.

Why this matters when you’re choosing a skill

Verification doesn’t guarantee quality. But it does guarantee accountability.

When a skill has a real source, you can see who wrote it and when. You can read the full instruction file before using it. You can check the commit history to see how it’s changed. You can decide whether the original author’s judgment matches yours.

A fabricated entry can’t offer any of that.

Source also tells you something about lineage. A skill from Anthropic’s official repository has a different trust profile than one from an account created last month with no other projects. You can see that distinction when you can see the source. You can’t see it when the entry was generated.

What you can verify on every skill page

• Repo owner — the GitHub account that published the skill
• GitHub folder — the exact folder within that repo
• Raw SKILL.md link — the full instruction file, one click away
• Upstream path — the canonical path used for installation
• License — MIT or Apache 2.0, confirmed from the source repo

Click the raw SKILL.md link. Read it. That’s the actual product you’re installing. The description on the skill page is a summary. The file is the truth.

What we don’t list

Skills with no associated GitHub repository. Skills with a repo but no structured SKILL.md. Skills that appear auto-generated with no real author. Skills with broken or redirected upstream links.

The directory stays smaller. It stays honest. Those two things are connected.

Where to start

Browse by what you do rather than by name. Filter by audience — developer, marketer, founder, writer — and you’ll see the skills most likely to match your actual work.

Or start with Skill Stacks, which groups verified skills into curated combinations for specific workflows. If you’re new to skills, a curated stack is faster than assembling your own collection from scratch.