October 30, 2026 · 6 min read
Claude Agent Skills for Brokers: Teach Claude Your Listing Checklist Once
An Agent Skill is a folder that teaches Claude one job. Here is what Anthropic says about them, and six steps to package a brokerage's listing launch checklist, voice guide and disclosure line as one.
An Anthropic Agent Skill is a folder with a SKILL.md file that teaches Claude one task, plus any reference files it needs. A brokerage can package its listing launch checklist as one: the steps, the voice guide and the disclosure line, written once, so Claude follows the office's process every time without anyone retyping it.
The checklist lives in a PDF. Three agents ask Claude for a launch set and get three different sets, one missing the disclosure line. The process exists, but only in people's heads, so the AI cannot follow it.
This post is part of our guide to AI for real estate marketing. It covers one question: what a skill is and how to build one from a checklist you already have. We use only what Anthropic's pages say about skills, checked on October 23, 2026. How that fits a brokerage is our reading. Menus and limits change, so confirm them first.
What a skill is, in Anthropic's words
Anthropic's engineering post on skills describes them as organized folders of instructions, scripts and resources that agents can discover and load dynamically. Its announcement about donating the standard puts it plainly: a skill is a folder containing a SKILL.md file that teaches an agent a specific task or domain.
Anthropic's developer documentation compares skills to an onboarding guide you would write for a new team member. A skill is that guide, written so Claude can read it. It loads in three stages.
The label. A name and a description. Claude reads only these at startup, and uses the description to decide when the skill applies.
The instructions. When your request matches, Claude reads the body of SKILL.md: the steps and rules.
The extras. Other files, such as a voice guide or sample posts, load only when the instructions point to them.
Skill or prompt: which one fits
Anthropic draws the line itself. Prompts are conversation-level instructions for one-off tasks. Skills load on demand, so you do not repeat the same guidance in every conversation. Rewriting one email is a prompt. A repeated job with steps is a skill, and our free Listing Launch Checklist is a good test case. It runs five phases in order: coming soon, just listed, first week, under contract and sold.
What goes in a listing checklist skill
A skill works when four things each sit in their own place.
The steps, in order. In SKILL.md. Short and numbered.
The facts rule. Claude drafts only from facts the agent supplies and the listing page confirms. If the price or listing agent is missing, it stops and says so. See our facts-only check for AI listing descriptions.
The voice. A separate file. No voice guide yet? Read how a voice guide keeps AI copy sounding like you.
The refusals. Describe the home and the offer, never who should or should not live there. No promises of results. No invented numbers. A line saying where a person checks before anything is posted.
Build it in six steps
Pick one job. Not "all our marketing." The listing launch set. A skill that does one thing is easy to trigger and easy to fix.
Write the check first. Anthropic's engineering post says to identify the gaps by running the agent on representative tasks. Pull five past listings and write what a correct launch set contained for each. That is your test.
Write the description. Say what the skill does and when to use it. Anthropic's help article on custom skills calls the description critical, because Claude uses it to decide when to invoke the skill.
Write SKILL.md. Frontmatter, then the numbered steps. Keep it short. When it grows, split it into separate files and point to them.
Zip the folder and add it. The help article says the ZIP should hold the skill folder at its root, not inside another folder. Then enable it under Customize and Skills in Claude.
Run your five test listings. Compare each result with the list from step 2. Fix the instructions, not the output, until all five pass.
Names allow 64 characters of lowercase letters, numbers and hyphens, and bar the words "anthropic" and "claude," per the developer documentation. So `listing-launch-checklist` works and `claude-listing-helper` does not.
A worked example: a 14-agent boutique brokerage
This office is invented. The broker has a one-page launch checklist and a two-page voice guide. She builds one folder, `listing-launch-checklist`, with four files.
SKILL.md. The label and the numbered steps.
voice-guide.md. The voice guide.
disclosure-and-fair-housing.md. The disclosure line and what never gets said.
approved-launch-sets.md. Two past sets she approved, as examples.
The description reads: "Runs our listing launch checklist for a new listing. Use when an agent sends an address, price and photos and asks for launch posts, the property email or the checklist." The body has six plain steps: confirm the address, price, listing agent, photo link and open house date; stop if the price or agent is missing; draft the carousel text, Facebook caption and property email from supplied facts only; apply the voice guide; add the disclosure line; end with a list of claims a person must verify against the listing page.
An agent types: "New listing, here is the address, price and photos." Claude matches the description, reads SKILL.md and asks for the missing open house date. The agent then gets a set shaped like the ones the broker approved, disclosure line included, and still checks it. The skill's job is a consistent first draft, not a published post.
By the numbers
Anthropic reports that more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, have integrated Agent Skills since it published the standard in December 2025.
The same announcement says 1 million skills have been published and nearly 280 million installed.
The developer documentation puts the always-loaded label at about 100 tokens per skill and the triggered instructions at under 5k tokens.
The description limit is stated two ways: 200 characters in the help article, 1,024 in the developer documentation. Write to the lower number and both are satisfied. The example above is 169.
None of this shows how well a skill performs for a brokerage. For that, run your own five listings.
Limits and risks to plan for
Who can use it. The developer documentation says custom skills on claude.ai are individual to each user, and do not sync between products. The help article has a note for Team and Enterprise plans that points to "Provision and manage skills for your organization." Anthropic's pages differ, so read the current help article before you tell forty agents it is available. Custom skills also need code execution enabled.
Where it came from. The documentation says to use skills only from trusted sources, meaning ones you wrote or got from Anthropic. A skill can include code, so have someone read every file before it goes to the team.
What it must not hold. Your process, not client data. Keep seller names and offers out.
Who checks. A person verifies every claim against the listing page before anything is posted.
How Urban Marketing Edge handles this
We use AI to move faster on drafts and variations, and we keep strategy and taste human. The idea under a skill is the one under our own work: write the steps once, so every request follows them. It is how we think about the listing launch system we built for Rocks Realty.
On our plans, a person on our team checks each draft against the client's facts and voice, and it goes through the post approval system before anything is posted. If your checklist and voice guide are not written down yet, that is the first job, with or without AI. The plans page shows what we handle.
Questions brokers ask
What is an Anthropic Agent Skill?
A folder containing a SKILL.md file that teaches an agent a specific task or domain, along with any scripts or reference material it needs. Claude reads the name and description first and loads the rest only when your request matches.
Is a skill different from a prompt?
Yes. Anthropic describes prompts as instructions for one-off tasks, while skills load on demand so you do not repeat guidance. Use a prompt for a one-off and a skill for a job you repeat.
Can my whole office share one skill?
It depends on the product and the plan. The developer documentation says custom skills on claude.ai are individual to each user. The help article has a note for Team and Enterprise plans that points to provisioning skills for an organization. Check the current help article before you plan a rollout.
Do I need to code to build one?
No. A basic skill is a folder with a text file and a few reference files, then a ZIP. Scripts are optional. A checklist, a voice guide and a disclosure line are plain writing.
Is it safe to download skills from the internet?
Anthropic says to use skills only from trusted sources, meaning ones you wrote or got from Anthropic. A skill can direct Claude to run code or call tools, so read every file before installing one.
Does a skill replace the person who checks the posts?
No. It makes the first draft consistent. A person still checks every fact against the listing page and approves the post.
Book a 30-minute strategy call
If your checklist lives in a PDF and your voice guide in one person's head, bring both. In 30 minutes we map the simplest plan for your content, themes and monthly cadence, and mark which steps are worth writing down for AI first and which stay with a person. No pitch deck, no pressure. Book a 30-minute strategy call, or see what each plan covers on the plans page first.
