A neutral, vendor-independent index of what AI agents can do

One canonical page per skill, regardless of platform. Each page answers four questions with no clicks: what the skill does, what runs it, how to install it, and whether it is any good. That last one is the differentiator — every listing carries a proprietary skill score derived from static code analysis, repository signals, and a documented rubric, so you can compare a Claude Skill, an MCP server, and a Cursor rule against each other on the same axis.

Pages are generated from a database, not hand-written. Forks and near-duplicates attach to a canonical entry instead of getting their own pages. Skills that fail the quality bar stay noindex until they earn indexing. This is deliberate — a directory padded with thin or duplicate pages gets a sitewide quality penalty and stops being useful.

• Not a vendor — we don't ship skills ourselves.
• Not a marketplace — paid skills are surfaced where they exist; the directory takes nothing.
• Not a review site driven by editorial opinion — the skill score is reproducible from public signals.
• Not pay-to-play — vendors can sponsor a clearly labeled placement next to organic results, never a better position in the organic results themselves.

Coverage across platforms is the moat. Every page is generated from the same database and the same template; nothing about the underlying vendor changes what a skill page looks like or where it ranks. The quality engine reads the same signals — code, documentation, repository activity, community ratings — regardless of who shipped the skill or how big their company is.

When a vendor's own skill ranks lower than a community alternative, that's the directory working as designed. When a vendor sponsors a placement, that placement is labeled and sits outside the organic results.

Today we cover Claude Skills, MCP servers, Cursor rules, Windsurf rules, Custom GPTs, n8n templates, LangChain and LlamaIndex packs, agent and system prompts, GitHub Copilot Extensions, Zapier and Make automations. The taxonomy is meant to grow with the ecosystem — as new platforms become widely used, we add them; old ones that fall out of use drop. Every category, every role, every industry, and every platform is browseable as a listing page.

If you maintain a skill we haven't indexed, or we have it but the metadata is wrong, submit it through the form — the same pipeline that crawls public repositories accepts vendor submissions and gives them a verified-author badge. If you've installed a skill from here, leaving an honest rating helps the next person find it (or skip it) faster.

The trust-score methodology is documented in full and the source for the scoring engine is open. If you think a particular score is wrong, the methodology page explains how to make the case.