Format guide

SKILL.md explained

Understand the role of instructions, metadata, supporting files, and source evidence—and what the format cannot prove by itself.

Last reviewed2026-07-30
Reading time7 min
Editorial ownerSkillSignal editorial
EvidenceSourceComputedTestedEditorial
Direct answer

SKILL.md is a source document that tells an agent how to approach a reusable task. Its value comes from clear scope, specific instructions, useful examples, and supporting assets. The file alone does not prove safety, successful installation, compatibility, or task quality.

What SKILL.md is for

A Skill packages repeatable domain instructions so an agent can apply a consistent workflow. A useful source file names the job, explains when to use it, defines steps, and makes expected outputs visible.

The best Skills reduce ambiguity without pretending every task can be solved by a rigid template.

Anatomy of a useful Skill

Metadata helps discovery and routing. Instructions shape the workflow. Supporting files can provide scripts, templates, examples, or reference material.

  • A specific name and description.
  • Clear triggers and situations where the Skill should not be used.
  • Inputs, workflow stages, and expected outputs.
  • Examples and validation criteria.
  • Explicit dependencies, permissions, and limitations.

How to evaluate source quality

Length is not quality. A short, specific Skill can be more useful than a long generic document. Evaluate whether each instruction changes how the agent performs the task.

Look for concrete decision rules, named outputs, domain constraints, examples, and a way to verify completion.

Computed quality is a screen

Documentation depth and specificity help prioritize review, but they do not replace a real task test.

What the file cannot prove

A polished source file cannot prove that linked scripts are safe, that installation succeeds, or that a platform loads every supporting asset as intended.

Claims about compatibility, security, or effectiveness need evidence beyond the document itself.

Source is evidence—not certification

Always inspect commands, dependencies, file access, network behavior, ownership, and the pinned commit.

Frequently asked questions

Is every Markdown instruction file an Agent Skill?

No. A Skill should have a clear reusable job, routing context, and enough structure to improve agent behavior. Generic notes may not provide that.

Can one Skill support several platforms?

The workflow may be portable, but installation and loading behavior vary. Treat multi-platform support as an evidence question.

Skills referenced in this guide

Computed 6982,073

nexu-io/open-design

8-bit-orbit-video-template

Hyperframes-based video template for retro pixel deck motion design. Use when users want a high-fidelity, multi-scene HTML-to-video composition with advanced transitions, interactive preview controls, and ready-to-render default style.

Computed 6635

asimons81/hermes-a2a-bridge

a2a-bridge

Discover and call text or structured JSON A2A agents through the Hermes A2A Bridge.

Computed 10042,015

coreyhaines31/marketingskills

ab-testing

When the user wants to plan, design, or implement an A/B test or experiment, or build a growth experimentation program. Also use when the user mentions "A/B test," "split test," "experiment," "test this change," "variant copy," "multivariate test," "hypothesis," "should I test this," "which version is better," "test two versions," "statistical significance," "how long should I run this test," "growth experiments," "experiment velocity," "experiment backlog," "ICE score," "experimentation program

Computed 8315

Servosity/msp-skills

abnormal

The full Abnormal Security REST API as an agent-ready CLI - with a local threat store, ranked SOC triage, and one-shot reporting no SOAR pack offers. Trigger phrases: `triage abnormal threats`, `remediate a phishing campaign in abnormal`, `abnormal email threat report`, `check account takeover risk for an employee`, `vendor email compromise check`, `use abnormal security`, `run abnormal-cli`.