How a Glencoe Wealth Advisor Can Build an AI Playbook With Claude Skills

Glencoe, Ill. , April 23, 2026. Park Avenue is quiet at 7:14 on a Tuesday morning, the way Glencoe is quiet most mornings before the 7:32 inbound to Ogilvie. The lights are on at the second-story office above the bookstore. A founding partner of a small wealth-advisory firm is at his desk with a coffee, a printed client portfolio, and a three-ring binder. The binder is labeled "Q2 review prep, Mr. and Mrs. K." It exists because the partner has done a Q2 review the same way for fourteen years and the way is in his head.

The associate who joined the firm in November is also at her desk. She has the same client. She has half the binder. She does not have the part of the binder that lives in the partner's head.

This is the problem Claude Skills, in their 2026 form, are built to solve.

If you run a small registered investment advisor on the North Shore, this article is about a specific operational change you can make in the next two weeks. It is not about a new chatbot. It is not about ripping out Wealthbox or Orion. It is about taking the playbooks that already exist inside your firm, the way you do quarterly reviews, the way you onboard a household, the way you tax-plan in October, and making them into reusable commands that any team member, including you, can run identically every time.

Glencoe's 2022 median household income was $228,750, per the village's most recent Census reporting. That is the math that makes a small Glencoe RIA work in the first place: a hundred to four hundred households, average AUM per household well into the seven figures, and a partner who personally knows every client. The model is excellent at the relationship and operationally fragile at almost everything else.

What Actually Changed With Claude Skills in 2026

For most of the last two years, the Claude ecosystem had two parallel ways to extend the assistant. Skills were folders containing a SKILL.md file. Slash commands were single Markdown files in a different folder. They did similar things, in subtly different ways, and you had to know which one to use when.

That distinction is gone. Per Anthropic's current Claude Code documentation: "Custom commands have been merged into skills. A file at .claude/commands/deploy.md and a skill at .claude/skills/deploy/SKILL.md both create /deploy and work the same way." The name field at the top of a SKILL.md file becomes the slash command. The description field controls when, and whether, Claude can pick the skill up automatically.

Three other things matter for an RIA principal who is not a developer.

Skills can auto-invoke. The default behavior, per Anthropic's docs, is that the description field is "always in context" and Claude loads the full skill the moment the conversation matches. If you describe a skill as "Use this when preparing a quarterly client review," and someone asks Claude to help them prepare a quarterly review, the skill fires. No slash typing required.

Skills can be locked to manual-only. Adding disable-model-invocation: true in the frontmatter forces the skill to wait for a human to type the slash command. Anthropic specifically recommends this for "/commit, /deploy, or /send-slack-message." For a wealth-advisory firm, that translates to anything that ends in client contact: an outbound email, a calendar invite, a filed compliance note, a journal entry on the custodian. You do not want Claude deciding on its own that it is time to email Mrs. K.

Skills can chain. Skills can call other skills. Mindstudio's coverage describes the architecture as having two layers: "Automation Layer: Skills handle repeatable, predictable work. Control Layer: Commands enable oversight without micromanagement." A practical version of this for an RIA is an orchestrator skill called /quarterly-review-all that runs the per-client /quarterly-review skill across a client list and stops for human review at the end. Anthropic's own example, context: fork, lets a skill spawn a forked subagent so the orchestrator's context stays clean.

Translate that for a small RIA: a Skill is where your firm's policies, your preferred talking points, your suitability questions, your compliance checklist, and your email tone live, in one file Claude reads every time the work shows up. Today most of that lives across three people, two notebooks, and a shared drive nobody opens. A Skill is the artifact that makes the firm survive the partner being on a sailboat in Croatia for two weeks in July.

Why This Specifically Matters for a Glencoe RIA

The Bace blog has covered wealth advisors using Claude to do more without hiring earlier this month, and there is a separate piece in circulation about Projects and Skills as building blocks. This article is a different angle: the playbook angle. The argument is not "use AI to save time." The argument is "the workflows you have already perfected over fifteen years are intellectual property, and the 2026 Skills system finally gives you somewhere to put them."

A Glencoe RIA has a particular operational profile. Most of the firms operating between Park Avenue and Vernon look the same from the outside. One or two founding partners. One or two associates, often a recent finance grad from Northwestern or Loyola. A part-time operations or compliance coordinator. A book of two hundred to four hundred households, almost all of them within a five-mile radius. Custody at Schwab or Fidelity. CRM in Wealthbox or Redtail. Performance reporting in Orion or Black Diamond.

The bottleneck is never tooling. The bottleneck is partner attention. The partner does the quarterly reviews because the partner is the relationship. The partner does the new-client intake because the partner is the trust. The partner does the October tax-planning sweep because the partner has been doing it since 2009 and remembers what each client did the last three Octobers.

The associates and the operations coordinator can do all of those things, in theory. In practice they cannot, because the playbook is not written down. Skills are the format the playbook gets written in.

One more piece of context. A small RIA does not have, and should not pretend to have, an engineering team. A skill is a Markdown file. The author is the partner who knows the workflow. The first draft is written in plain English. The whole thing fits on one screen. There is nothing to deploy and nothing to maintain beyond a quarterly review of whether the file still reflects how the firm actually works.

Use Case 1: The /quarterly-review Skill

The single most expensive recurring meeting in a small RIA is a quarterly review the partner had to prep alone at 6 AM the morning of.

Walk through what a "good" quarterly review prep looks like at most Glencoe firms. The partner pulls the household's portfolio out of Orion. The partner reads back through the meeting notes from the last review. The partner checks for any household life-events recorded in Wealthbox in the last ninety days, a college graduation, a job change, a new grandchild. The partner glances at any flagged emails from the household since the last touch. The partner drafts three to five talking points: performance versus benchmark, allocation drift, planning topics for this quarter, an item or two specific to the household. The partner writes a short pre-meeting email confirming the time and previewing the agenda.

That whole sequence takes a good partner forty-five to seventy-five minutes per household. At a 250-household book reviewed quarterly, the partner is spending twenty to thirty hours per quarter just on review prep, not the meetings themselves.

The Skill version replaces that.

That file goes in .claude/skills/quarterly-review/SKILL.md. It is around 200 words. It will become the /quarterly-review command. The associate types /quarterly-review Mr. and Mrs. K. and gets the packet. The partner reviews the packet, marks it up, and walks into the meeting an hour after starting instead of three hours after starting.

Two design choices matter. First, the description starts with "Use when the partner asks to prep a Q1, Q2, Q3, or Q4 review." That phrasing lets Claude auto-fire the skill if anyone on the team asks a question that matches the pattern, even if they forgot the slash command. Second, step six explicitly says "Stop. Do not send anything." A wealth-advisory skill should almost never autonomously send client communication. The Skill format makes that boundary explicit.

For a firm that wants to go further, the same skill can be set to run as a context: fork subagent, so each household's prep happens in an isolated context, and an orchestrator skill called /quarterly-review-batch runs the per-household skill across the next two weeks of scheduled meetings. The partner walks into Monday with twelve packets pre-built, sips coffee, marks up, and starts the week.

Use Case 2: The /new-client-intake Skill

Onboarding a new household is the single most-handled-by-the-partner workflow in a small RIA, and the single most worth standardizing.

A new-client intake at a small RIA touches roughly fourteen artifacts in the first thirty days: a discovery questionnaire, a risk-tolerance instrument, a suitability narrative, a fee disclosure, an ADV delivery, a privacy notice, a custodial application, a beneficiary form, a transfer authorization, a CRM record, an Orion household setup, a welcome packet, a thank-you note, and a calendar invite for the first follow-up. Each of those exists somewhere in the firm. They do not exist in one place, in one order, with one owner.

The Skill version pulls the whole sequence into one file. The partner who built the firm knows how the sequence should run. The Skill is where that knowledge becomes the firm's standard operating procedure, in a format any team member can execute.

Notice the disable-model-invocation: true line. Onboarding is exactly the kind of workflow that should never run on its own. It only runs when someone on the team types /new-client-intake "the K. Household". Per Anthropic's docs, that flag means "the description not in context, full skill loads when you invoke." It removes any chance Claude offers to "help" run an intake when nobody asked.

The output is a packet. The packet is reviewed by the partner and executed by the operations coordinator. The associate, who used to spend a half day per intake chasing the partner for the next step, instead spends ten minutes reviewing the Skill output, queuing the documents in DocuSign, and entering the CRM fields. Time-to-onboard goes from two to three weeks to roughly five business days.

This is also where the firm's intellectual property accumulates. The first version of the skill is whatever the partner currently does. Six months in, when the partner notices that one suitability question keeps surfacing useful information, they edit the SKILL.md file and the firm's intake gets quietly better. There is no software release. There is just a Markdown file the partner edited at the kitchen table on a Sunday.

Use Case 3: The /annual-tax-planning Skill

October and November in a small RIA is a six-week sprint of tax-loss harvesting and RMD reminders, run from a partner's memory of last year's spreadsheet.

Every fall, a small RIA does the same sequence of work for every household. Identify positions with embedded losses for tax-loss harvesting. Confirm or initiate required minimum distributions for households over the RMD age. Check for charitable-giving plans, especially qualified charitable distributions for IRA holders over 70 and a half. Coordinate with the household's CPA on year-end estimated payments. Flag any Roth conversion windows. Send a short year-end letter.

The work itself is repetitive. The volume is what makes it brutal. At a 250-household firm, that is 250 households to walk through in October and November. The partner cannot do it alone. The associate cannot do it alone. The firm muddles through, every household gets touched, but the consistency varies wildly: clients with simple situations get treated identically, clients with complicated situations get a deep custom treatment, and clients in the middle get whatever the team had time for.

A Skill changes the math.

That skill is the play. Run /annual-tax-planning Mr. and Mrs. K. for every household over the first two weeks of October. The associate handles execution. The partner reviews each one in the morning, signs off, and the operations coordinator routes the client letters and CPA notes for the partner's signature.

The firm's actual planning intelligence, which positions are on the approved-substitute list, what counts as "embedded losses worth harvesting," what the RMD logic should do for a household with multiple inherited IRAs, lives inside the Skill. As the firm refines the rules, the file is edited. The associate two years from now inherits a fully-codified play, not a verbal walk-through from a partner who left.

One more pattern worth noting. Nothing prevents the firm from writing an orchestrator skill called /annual-tax-planning-all that runs the per-household skill across the entire book and produces a summary spreadsheet for the partner. Per Anthropic's documentation, this is the chaining pattern: "Skills can call other skills." The orchestrator runs every weekday morning in October, the partner gets a coffee and a list of the ten most consequential households to look at first, and the rest of the book proceeds in a steady cadence the associate manages.

How to Build Your First Three Skills in Two Weeks

The mistake every firm makes in week one is trying to write skills in committee. The right sequence is much smaller. Pick one workflow. Write the skill in plain English. Test it on three real clients. Iterate. Then move to the next workflow.

What This Does Not Replace

One direct note, because it matters. A Skill is not a substitute for the partner's judgment, the partner's relationship, or the partner's licenses. The Skill prepares the packet. The partner reads the packet, decides what to recommend, and conducts the meeting. The Skill drafts the email. The partner edits the email and signs it. The Skill produces the tax-planning summary. The partner is responsible for what gets executed.

This is doubly true for anything compliance-adjacent. A Skill that "drafts" an ADV delivery is fine. A Skill that "files" anything with regulators is not. The firm's compliance officer, internal or outsourced, still owns every piece of output that leaves the building. A Skill makes the workflow faster and more consistent. It does not change who is accountable for it.

The other thing a Skill does not replace: the partner's ear. The reason a small Glencoe RIA wins against the big wirehouses down in the Loop is that the partner notices when something in a quarterly review feels off, when the client's voice in the meeting changed, when the question about long-term-care insurance is really a question about a parent in cognitive decline. A Skill cannot do that. It can free up the partner's calendar and attention so the partner is in a position to do it.

A Final Note on Timing

Skills as the unified slash-command system are new in 2026. The next twelve to eighteen months are the window during which Glencoe RIAs that build a real internal playbook will quietly pull ahead of the firms that do not. The advantage is not visible in any single quarter. It compounds.

By 2028, the firms that did this will have associates who run quarterly reviews to a standard the partner approves of, intake processes that survive the partner being out, and tax-planning sweeps that touch every household with the same rigor as the partner's favorite five. The firms that did not will still be doing reviews at 6 AM the morning of.

For a partner who wants a second set of eyes on which workflow to write up first, or how to set the firm up to actually use the file once it exists, a free 30-minute AI audit is available. In person in Glencoe or on video. No obligation, no pitch deck. The output is a one-page plan listing the three workflows worth turning into Skills first, and what each one would save the firm in partner hours over the next quarter.

The Skill format is plain Markdown. The hardest part is deciding which workflow to write down first.

Frequently Asked Questions