Methodology

How this reference is built

This page exists so that a reader can check the reference instead of trusting it. It says where each kind of claim comes from, how the tools on the site work, and what none of it can prove.

Where the format rules come from

Every rule about SKILL.md, from the name format to the description length, is taken from the published Agent Skills specification and linked to it. Where the specification calls a field experimental, the page says so. Statements about how Claude's products load or share skills come from Anthropic's own documentation, and the changelog records the day it was read.

What was read for published skills

The conflicts and pairings page is written from the skills' own published instructions. A conflict is recorded only when two skills instruct the agent on the same decision, not when they merely cover the same subject. Star counts and install counts are not published, because they move daily and cannot be re-verified on every build.

How the tools work

The description checker and the skill generator run entirely in your browser. The checker applies the specification's hard limits first, then a few heuristics: whether the description says when to use the skill, and whether it is specific enough to match a task. Passing means the text is well formed, not that an agent will pick the skill for every task.

What this does not prove

Product behaviour changes faster than the specification, so the changelog records each date a source was re-read and each correction. None of this is a security audit. Read a skill and its scripts before you install it into an agent that can run commands on your machine.