Claude Cowork Skills Explained: What They Are and How They Work

Right now you're probably rewriting the same prompt every time, or copying it from somewhere and pasting it in. Skills fix that write your workflow once, run it with a single command. What we cover: → What a skill actually is: a plain markdown file with a description and step-by-step instructions → Live demo: video metadata skill generates 3 title options, a description, and a thumbnail from a transcript with one command → Why the description is the most critical part of any skill it's how Claude auto-selects the right skill without you naming it → The filing cabinet analogy: skills only load what's needed, so your context window stays clean → The 3 layers of a skill: metadata (always in memory), body (loads when triggered), reference files (on demand) → How 100 skills only use ~10,000 tokens of context everything else waits until called This is Part 1 of the Claude Skills series. Timestamps: 00:00 — What skills are and why they matter 00:39 — The problem skills solve: repetitive prompts 00:56 — Live demo: video metadata skill from transcript 02:16 — Why skill context showing in the session matters 04:11 — What a skill is: plain markdown text file 04:44 — The description: the most critical part of a skill 05:20 — Demo: skill auto-loads without naming it 06:51 — Skills as documentation write once, reuse forever 07:10 — The filing cabinet analogy explained 08:40 — 3 layers of a skill overview 09:14 — Layer 1: metadata (always in memory) 09:51 — Layer 2: skill body (loads when triggered) 10:29 — Layer 3: reference files (on demand) 11:21 — Progressive disclosure: 100 skills, minimal context cost