Which Claude Skills Actually Run in Production?
If you have already looked at the open-source Claude skills repo for real estate agents, you have probably seen the full list of documented procedures. The real question — the one agents ask us every week — is not ‘what is in the repo?’ It is ‘which of these actually runs in a live production business, and what does the human still have to do?’ That is what this post answers. We are going past the overview and into the mechanics of which skills do real work, where they hand off to the licensed agent, and why that split is not a limitation — it is the whole design.
Quick answer: The skills that run consistently in production are the ones that live entirely on the public-records side of the workflow — permits, deed checks, property records, packet assembly, listing remarks, and CRM nurture drafts. None of them touch the MLS. None of them select comps. None of them produce an opinion of value. The agent does that. The skills do the assembly so the agent is not spending three hours chasing county portals and formatting packets before the real work begins.
The Split That Protects Your License — and Why It Is the Whole Design
Before we go skill by skill, you need to understand the architectural decision behind the open-source Claude skills repo for real estate agents, because it is not accidental. Every skill in the repo was built around one principle: the licensed human runs the licensed work. The automation runs the public-records assembly. Those two halves combine into a finished product. Neither does the other’s job.
In practice, that means this: you run your MLS. You export your own comps. You choose the properties that belong in the analysis. You apply adjustments. You sign the opinion of value. That is your license, your liability, and your professional judgment — none of which a documented skill should be near.
What the skills handle is everything you were doing by hand that did not require a license to do: pulling permit records from county portals, cross-checking deed restrictions in the Register of Deeds, assembling a pricing packet from the data you already chose, drafting listing remarks in your established voice, building CRM nurture sequences from the lead profile you already reviewed.
NAR Code of Ethics Article 11 and state license law put the opinion of value on the licensee personally. The automation organizes what the agent already decided. Time is saved on assembly and formatting — never on analysis. Build your skills around that line and you will never have a compliance problem with your workflow. Blur that line and you have invented a liability.
For the full documented list of what the repo covers, visit the open-source Claude skills repo for real estate agents — that page is the canonical source. This post goes narrower: specifically which of those skills we run in production every day and what the agent-side inputs look like.
The Permit and Public Records Skills — Where the Real Time Savings Are
If you had to pick one category where documented Claude skills return the most time per use, it is permits and public records. Here is why: every production agent already knows they need this information. The question is whether you spend forty-five minutes chasing county portals yourself, or whether a documented skill does that run while you are pulling your MLS data.
What the Permit Skill Actually Does
The permit pull skill checks county permit portals and open data sources for the subject property. It looks for additions, conversions, and accessory structures — and critically, it flags whether the certificate of occupancy is closed or open. An open CO on an addition the seller is advertising as finished square footage is a material fact issue. You want to know that before the pricing conversation, not during the inspection.
What it does not do: it does not interpret whether that open CO affects value. You make that call. The skill surfaces the fact; you decide what to do with it.
The Deed and Ownership Check Skills
The /deed-check and /ownership-check skills pull from Register of Deeds data and county GIS parcel records. They look for covenants, amendments, owner-occupancy restrictions, and leasing restrictions. If a buyer is planning to convert a property into a short-term rental and there is a covenant restriction on the deed that prohibits it, that is something that should surface in the research phase — not after the purchase agreement is signed.
Again, you review. You decide. You disclose. The skill does the chase so your review starts from a complete file instead of a blank page.
Listing Remarks and the Voice Problem Every Agent Has
Listing remarks are one of the highest-leverage places to deploy a documented Claude skill — and one of the most misused. The mistake agents make is treating listing remarks like a one-time prompt: they describe the property in a chat window, get a draft, clean it up, and move on. That works once. It does not scale, and it does not maintain voice consistency across your listings.
How the /listing-remarks Skill Solves This
The /listing-remarks skill in the repo is a documented procedure with a defined input structure. You feed it your established facts: square footage from your licensed measurement or a verified source, confirmed bedroom and bathroom count, lot size, feature list from your walkthrough notes, and any details the seller emphasized as selling points. The skill drafts remarks in a voice that has been documented — your voice, your market’s vocabulary, your compliance language.
One critical note on square footage: tax card square footage is noted by the skill but never used as the pricing anchor. Appraisers measure to ANSI Z765. Your skill should reflect that discipline. If your listed square footage cannot be verified by a licensed measurement or a confirmed source, the skill flags it rather than using an unverified figure.
Fair Housing Compliance Is Built Into the Skill Structure
A properly documented listing remarks skill includes compliance checkpoints that a raw prompt does not. Every output gets reviewed against a Fair Housing language check before it is used. The skill does not drift into protected-class language because the procedure explicitly prohibits it — and because you review before anything publishes. That review step is not optional. It is part of the skill’s defined workflow.
CRM Nurture Skills That Actually Get Used — and the Ones That Do Not
The CRM nurture category is where agents have the most uneven results with documented skills — not because the skills are weak, but because agents skip the input discipline and wonder why the output does not convert.
What Makes a CRM Nurture Skill Work
The /agent10_crm_nurture workflow runs on a schedule in our production system — not on demand. That distinction matters. A nurture skill that only runs when you remember to trigger it is not a nurture system. It is a better prompt. The value of a documented skill is that it runs at a defined time, against a defined contact list, and produces a defined output format you can review and deploy without rebuilding the context every time.
The input side is where most agents underinvest: the skill is only as useful as the lead profile you feed it. If your lead data is thin — just a name and phone number — the nurture draft will be generic. If your lead data includes source, property interest, timeline, last touchpoint, and any notes from prior conversations, the draft will be specific enough to actually send.
The Skill That Enriches Before the Nurture Runs
The /enrich-leads skill and the enrich_leads_pipeline.sh cron handle lead enrichment before the nurture workflow fires. Public records, property history, and any available profile data get pulled and added to the contact record. The nurture draft then has something real to reference. That sequence — enrich first, then nurture — is what separates a CRM skill that produces usable output from one that produces a template you immediately rewrite.
What Breaks When Agents Skip the Human-AI Split
We have had conversations with agents who came to us after attempting to build AI workflows that crossed the line we described above. The patterns are consistent and worth naming directly so you can avoid them.
Attempting to Automate the Comp Selection
This feels like a natural step — you have the data, Claude is good at analysis, why not let it pick? Here is why not: comp selection is a professional judgment call that belongs to the licensed agent. It is not a data-sorting problem. Two properties that look identical in a spreadsheet can require completely different adjustments based on conditions you observed during a walkthrough. A skill that selects comps is not saving you time — it is producing an output you cannot trust and cannot sign.
The correct workflow: you pull the comps, you choose the properties, you export your selection. The skill then runs the math against your choices, assembles the packet, and formats the output for your review. That is the workflow that is both faster and defensible.
Using Unverified Square Footage as a Pricing Input
Several publicly available data sources — Redfin, Zillow, county tax cards — provide square footage figures that may not match a licensed ANSI Z765 measurement. A skill that uses an unverified tax card figure as its primary input for a pricing packet is building the wrong foundation. The correct skill design notes the public figure, flags any discrepancy between sources, and defers to the licensed measurement or the agent’s stated verified figure. If your workflow does not make this distinction, your packet is based on data you cannot stand behind.
Prompt vs. Skill: A Side-by-Side Comparison
Because this distinction is central to the value of the skills repo, here is a direct comparison of how a raw prompt and a documented Claude skill differ across the dimensions that matter in production:
| Dimension | Raw Prompt | Documented Claude Skill |
|---|---|---|
| Context retention | Resets every session | Voice, compliance rules, and workflow logic are documented in the skill definition |
| Consistency | Varies with how well you explain the context that day | Same structure and output format every run |
| Compliance checkpoints | Depends on what you remember to include | Built into the procedure definition — cannot be skipped |
| Scheduling | Runs when you remember to run it | Can be triggered by a cron on a defined schedule |
| Team scalability | Lives in your head; cannot be handed off | Documented; another agent on your team can run the same skill the same way |
| MLS access | Prompt has no MLS access; must be manually fed | Skill has no MLS access either — by design. Agent runs the MLS; skill reads the export |
| Licensing risk | Higher — no defined boundary on what the AI is asked to do | Lower — skill definition explicitly states what it does not do |
| Setup time | Instant, but repeated every session | Upfront documentation investment; zero setup time after that |
How to Get Started Without a Coding Background
The most common objection we hear when agents look at the repo is that they assume it requires coding knowledge to use. It does not — but it does require documentation discipline, which is a different skill that most agents already have and undervalue.
Step One: Document One Workflow You Already Do
Pick the task in your production week that is most repetitive, most time-consuming, and least judgment-dependent. For most agents, that is either permit research, listing remarks, or CRM follow-up drafts. Write down exactly what you do, step by step, including what inputs you start with and what output you need. That documentation is the skeleton of your first skill.
Step Two: Match It to an Existing Skill in the Repo
The open-source Claude skills repo for real estate agents already has documented procedures for the most common production workflows. Before you build from scratch, check whether a skill already exists that matches your workflow. If it does, you can adopt and modify it in far less time than building new. The MIT license means you can take it, adapt it to your market, and run it without restriction.
Step Three: Define Your Compliance Checkpoints Before You Run Anything
Every skill needs at least two compliance checkpoints defined before it touches production: one at the output stage (does this draft contain any Fair Housing proxy language?) and one at the handoff stage (did a licensed agent review this before it was used in a client-facing context?). Document those checkpoints in the skill definition, not as a reminder to yourself. A checkpoint that lives in your head is a checkpoint that gets skipped on a busy Thursday.
For agents who want a structured guide to building their first documented AI workflow, the CEO Time Audit is a useful starting point — it helps you identify exactly where your production hours are going before you decide which workflow to automate first. And the follow-up system guide covers the CRM nurture side of the workflow in depth if that is where your biggest time drain lives.
What Real Estate Agents Are Saying About AI Tool Choices in 2026
Industry headlines this week flagged over twenty AI tools competing for the attention of real estate agents. Most of them are point solutions: one tool for listing descriptions, one for social content, one for lead scoring. The challenge with point solutions is that they do not talk to each other, they each have their own pricing and compliance terms, and none of them are built on your voice or your data. A documented Claude skill runs on Claude Code, uses your exported data as its input, and produces output in your documented voice — no additional subscription, no third-party data sharing, no vendor lock-in.
That is not an argument against every AI tool. It is an argument for understanding what you are buying when you pay for a point solution versus what you build when you invest in a documented skill. The agents who are building skills are building leverage that compounds. The agents who are subscribing to tools are renting capability they do not own and cannot modify.
Why Al and Victoria Pinder Built This in Public
The open-source Claude skills repo for real estate agents was not built as a marketing asset. It was built because Al and Victoria Pinder were running these workflows in their own production business and realized that documenting them publicly was more useful than keeping them proprietary.
Al has been at eXp Realty since the beginning of his career — not as a transfer from another brokerage, but as an agent who built his entire production business inside this model. The workflow philosophy that shaped the skills repo came from that same production discipline: build systems that run on your data and your judgment, eliminate the manual assembly work, and invest the recovered time in the decisions that require a licensed professional.
In the early years, that meant working through the full cycle of paid lead platforms — a revenue-split deal, bought zip codes, then a six-month Zillow contract that produced no conversions. By year three, every paid lead platform was gone. Not because the platforms failed philosophically, but because they had built their own pipeline and the platforms were no longer adding value relative to their cost. That three-year arc is why every skill in the repo is built around the agent’s own data — because an agent who owns their CRM, their lead data, and their workflows does not need to rent someone else’s pipeline.
If you are ready to build your first documented Claude skill and want to talk through where to start in your specific production workflow, book a strategy call with Al and Victoria Pinder at calendly.com/theprosperityagent. Or grab the Blueprint to see the full framework: theprosperityagent.com/resources/blueprint/.
Frequently Asked Questions
What is the open-source Claude skills repo for real estate agents?
The open-source Claude skills repo for real estate agents is a free, MIT-licensed library of documented AI procedures built specifically for production real estate workflows. Each skill defines a specific input, a specific output, and compliance checkpoints. Skills cover permit pulls, deed checks, listing remarks, CRM nurture drafts, and packet assembly — none of them touch the MLS or produce an opinion of value, which remains the licensed agent’s responsibility.
Do Claude skills for real estate require coding knowledge to use?
No. Using an existing Claude skill from the repo requires documentation discipline, not coding. You need to understand the skill’s defined inputs, provide your own verified data, and review outputs before use. Building a new skill from scratch may involve light scripting if you want it to run on an automated schedule, but adopting and modifying an existing skill is accessible to any agent willing to invest in the upfront documentation work.
No — and this is by deliberate design. The skills repo does not include any skill that logs into an MLS, scrapes listing data, or selects comparables. The skill reads that export and handles the assembly side. This split is how the workflow stays compliant with MLS terms of service and state license law, which places the opinion of value on the licensed agent personally.
How is a documented Claude skill different from a raw AI prompt?
A raw prompt resets every session — you re-explain your market, voice, and compliance rules each time. A documented Claude skill has those elements baked into its procedure definition. It produces the same structured output format every run, can be triggered on a schedule, and can be handed off to another agent on your team without losing context. The upfront documentation investment is the only cost; after that, the skill runs without setup time.
What Claude skills run in a real production real estate business?
In active production, the skills that return the most consistent value are permit and public records pulls, deed and covenant checks, listing remarks drafting, and CRM nurture sequence generation. These are all public-records or assembly tasks that do not require a license to perform but were previously eating hours of agent time per transaction. Skills that attempt to cross into comp selection or pricing analysis are outside the design intent of the repo and create compliance risk.
Is the Claude skills repo actually free to use and modify?
Yes. The repo is MIT-licensed, which means you can use it, modify it for your market, and build on it without cost or permission. The only requirement under MIT is that you retain the license notice if you redistribute. Agents can adopt a skill, adapt the documented procedure to their specific county data sources or voice preferences, and run it in their own production workflow without restriction.
Where do Claude skills for real estate agents live and how do I install one?
The canonical source is the open-source Claude skills repo for real estate agents at theprosperityagent.com/skills. Each skill is a documented procedure file. Installation means placing the skill definition where Claude Code can reference it and providing your local data exports as inputs. The repo page includes setup guidance for agents without a development background.