RESOURCE LIVE

Write a skill once and stop re-explaining yourself.

A skill is a piece of your judgment written down where the agent reads it. Done properly they compose, they version, and they hold up months later. Done the usual way they drift, break quietly, and nobody touches them again. This is the difference.

Raj Singh By Raj Singh · August 2026 · 9 min read

A SKILL, IN THE WILDWhat the agent reads before it starts

SKILL.md · frontmatter

v2.1 · 14 uses this week

›name — deploy-check. Loads on any deploy or release request.

›triggers — Six phrasings, including the two people actually type.

›body — Nine steps, one rollback rule, three worked examples.

Written once. Read every time.Versioned · tested

THE TELL

You write a skill, it drifts, you never open it again.

→A skill is the difference between an agent that helps today and one that is better in three months. Almost nobody treats them that way.
→Most published skill content is a list of prompts. Nobody shows the architecture, so nobody's second skill fits alongside their first.
→Two skills that overlap will fight, and the agent will pick one at random. You find out from an output that is subtly wrong.
→Skills regress silently. Nothing tells you a skill stopped firing, it just quietly stops being used.

WHAT'S INSIDE

Anatomy, composition, versioning, tests.

01
The anatomy
Frontmatter, triggers, examples and version, and what each part does to whether the skill actually loads when it should.
02
Triggers that fire
Writing the description so the skill loads on the phrasing people use, not the phrasing you had in mind when you wrote it.
03
Composition
Structuring skills so they stack instead of competing, including the scoping rule that stops two from claiming the same job.
04
The seven to ship first
The foundational set that pays for itself fastest, in the order I would add them to a new setup.
05
Tests and versions
How to catch a skill that has silently stopped firing, and the versioning habit that makes a rollback one line.

BUILT FOR

You, if you keep re-typing the same correction.

If you find yourself telling an agent the same thing every week, that correction is a skill you have not written yet. Start with one, and the pattern is obvious after that.

Why now Skills only recently became worth engineering properly. Once agents began loading them reliably from a description alone, the design of that description turned into the whole game.

The compounding is the real thing. Every correction you write down once is a correction you never make again, and after a few months the agent argues with you less because it already knows the answer.

Person writing notes on paper at a desk — Photo: Hadis Malekie / Unsplash

Get the resource

Drop your email, get the full tool.

One email. Instant access. No sequence, no upsell, no list-trap. I'll only ping you if I ship something that fits.

One email. No follow-up unless I ship something that fits.

The kit: a Claude Code skill is frontmatter, triggers and examples. Get the anatomy right and it fires on the phrasing people actually use, stacks with your other skills instead of fighting them, and tells you when it silently stops working.

→skill-anatomy.md is the exact frontmatter shape — name, description, triggers, examples, version — and which field actually decides whether the skill loads.
→trigger-phrasing.md writes the description against how people actually ask, not the phrasing you had in your head, so the skill fires on the real request instead of sitting unused.
→composition-rules.md is the scoping pattern that lets skills stack instead of fight, so two overlapping skills don't leave the agent picking one at random.
→the-seven-to-ship-first.md is the foundational set that pays for itself fastest, in the order I'd add them to a new setup.

→Claude Code 1.0+ with skills enabled
→A skills directory (.claude/skills/) in the project or user config
→One real recurring task to encode — skills written from a hypothetical never fire right
→30 minutes for the first skill, once you're not guessing at the frontmatter shape

↑ Unlocks the moment you submit the form above.

WRITTEN BY

Raj Singh

I'm Raj Singh, founder of DebtOps.ai and a Claude certified partner. Over the last five years I've managed $25M in ads, built AI solutions for Fortune 100 companies, and installed AI operating systems and AI brains inside small businesses. Everything here is something I've shipped, debugged, or watched a client ship. My goal with AgileSingh is to make you irreplaceable in this AI age.

$25M

Ads managed · 5 yrs

F100

Enterprise AI builds

Claude

Certified partner

Guides shipped

Browse all resources · DebtOps.ai · @agilesingh on X

More guides

Now put the skills in front of a swarm.

Skills scale better than agents do, because twenty agents read the same file. The agent swarm guide is where that pays off, and the setup guide is the config underneath. Browse the library, or pin a slot and we will write your first three.

All resources Work with me