Locus product guide · Updated November 10, 2026

What are agent skills? Claude Skills and SKILL.md

Agent skills are folders of instructions, scripts and reference files that an AI agent loads only when a task needs them. Each skill has a SKILL.md file that starts with a name and a description. The agent keeps about 100 tokens of that metadata in context per skill and reads the rest on demand. Anthropic created the format and published it as an open standard at agentskills.io. Claude Skills use it in claude.ai, Claude Code and the Claude API.

What is inside an agent skill?

A skill is a directory. The only required file is SKILL.md: YAML frontmatter, then Markdown instructions. Optional folders hold scripts, reference docs and assets such as templates.

What is inside an agent skill?
Frontmatter field	Required	Rules in the open spec
name	Yes	Up to 64 characters. Lowercase letters, numbers and single hyphens. Must match the folder name.
description	Yes	Up to 1,024 characters. Says what the skill does and when to use it.
license	No	A license name or a bundled license file
compatibility	No	Up to 500 characters on required tools, packages or network
metadata	No	Any string keys and values, such as author and version
allowed-tools	No	Experimental. Tools the skill may run without asking.

How does an agent load a skill?

Skills load in stages, which the docs call progressive disclosure. That is why you can install many skills without filling the context window.

How does an agent load a skill?
Level	When it loads	Context cost
Metadata: name and description	At startup, for every skill	About 100 tokens per skill
Instructions: the SKILL.md body	When a request matches the description	Under 5,000 tokens recommended
Resources and scripts	Only when the instructions point to them	None until read; scripts add only their output

Where do Claude Skills work?

Claude Skills follow the open format. Each Claude surface adds and runs them differently, and custom skills uploaded to the Claude API do not show up on claude.ai, or the other way round.

Where do Claude Skills work?
Surface	How you add a skill	What it can reach
claude.ai	Upload a zip in settings. Anthropic’s document skills are built in.	Code execution must be on. Network access depends on admin settings.
Claude Code	A folder in ~/.claude/skills or .claude/skills, or a plugin	Full network access, like any program on your computer
Claude API	Upload through /v1/skills, then list skill IDs in the container parameter	Runs in the code execution sandbox with no network access

What can a skill do that a prompt cannot?

A prompt or custom instruction is always loaded. A project adds the same background to every chat in it. A skill loads only when the task matches, so you can keep dozens of them ready at little cost.

Skills can also bundle code. Claude runs a script with bash and sees only its output, not the code. That makes repeated steps faster and more consistent than asking the model to write the code each time.

Claude Code adds its own fields. disable-model-invocation: true makes a skill run only when you type /skill-name. user-invocable: false hides it from the menu so only Claude can use it. Uploads to claude.ai and the Skills API accept only the open-spec fields.

Are agent skills safe to install?

Treat a skill like software. Its instructions and scripts run with whatever access the agent has. Anthropic advises using skills only from sources you trust and auditing every file first.

Read SKILL.md and every script for network calls or file access that do not fit the stated purpose.
Be careful with skills that fetch content from URLs. The fetched text can carry new instructions.
Pre-approved tools in allowed-tools skip the usual prompt, so check that list.
Claude Enterprise can scan custom skills uploaded in claude.ai and Cowork. Scanning does not cover the Skills API.

What do the Locus Pro skills do?

Locus publishes an installer skill at https://paywithlocus.com/SKILL.md. Tell any agent: Install Locus Pro by following https://paywithlocus.com/SKILL.md. The agent picks MCP, the Locus CLI or plain HTTP, signs you in, and installs three skills: locus, locus-setup and locus-workflows.

The locus skill tells the agent when to reach for Locus: before it scrapes a site with curl or Python, when web tools fail, and for cited research, papers, audio, people and companies. It then explains how to search the catalog, read a tool’s price and run it. The calls themselves go through the Locus MCP server or CLI.

The skill costs nothing. The paid calls it leads to bill one prepaid credit balance, where 1,000 credits equal $1. Firecrawl scrape costs 1 to 38.6 credits and was 1 credit on most recent calls. Perplexity search costs up to 5.8 credits. Hunter email finder costs 1 to 10.1 credits. Prices checked November 10, 2026.

When should I write a skill?

Write a skill when the agent already has the tools but keeps doing a task the wrong way.

A repeated workflow, such as a release checklist or a report format: a skill.
Company knowledge, such as brand rules or a data schema: a skill with reference files.
A step that must be exact every time, such as filling a form: a skill with a script.
Access to a new system or live data: an MCP server, often paired with a skill that explains how to use it.

Skills vs MCP compared

Frequently asked questions

Are Claude Skills the same as agent skills?

Yes. Claude Skills follow the Agent Skills open standard. Claude Code adds extra frontmatter fields, such as invocation control and running a skill in a subagent, that other tools may ignore.

Do skills cost anything to use?

There is no fee for a skill. It uses context tokens when it loads, and any script runs in your code execution environment. Tools the skill tells the agent to call may charge on their own.

Which Claude plans include skills?

The Claude help center lists skills on Free, Pro, Max, Ultra, Team and Enterprise, with code execution turned on. Team and Enterprise owners can also provision skills for every member.

Can the same skill work in other agents?

Often, yes. The format is open and other agent products have adopted it. Check that the target agent supports any optional fields you use, such as allowed-tools.

Implementation references

Use these first-party references for current request contracts and account requirements. Tool availability, prices, and negotiated terms can change.

Sources

Third-party details were checked on November 10, 2026. Vendors change pricing and features often, so confirm on their site before you decide.