Claude is probably running an old copy of your skill. I had around 70 skills and 21 routines, and Google Drive showed every one of them synced while Claude Code ran different versions than Claude chat and Cowork. In this episode I install my free template from zero on a fresh account: one GitHub repo that Claude Code, Claude chat and Cowork all update from, with Vale (a free style checker) blocking a broken house rule instead of trusting Claude to remember it. If your skills live in more than one place, they have already drifted.

The free template, with the one line to paste into Claude Code:
https://github.com/serious-ai/claude-skills-starter

Matt Pocock's skills repo, where this started:
https://github.com/mattpocock/skills

Serious AI News, every Monday. AI for operators, what we built and what it cost:
https://seriousai.partners/newsletter/?utm_source=podcast&utm_medium=description&utm_content=ep-329-claude-skills-repo

Tell us what's breaking:
https://seriousai.partners/contact/?utm_source=podcast&utm_medium=description&utm_content=ep-329-claude-skills-repo

Chapters:
0:00 Claude is probably running an old copy of your skill
1:12 The growing pile of skills nobody can track
2:04 69 skills and 21 routines in one GitHub repo
3:07 Drive said synced, Claude ran other versions
4:38 Where it started, Matt Pocock's skills repo
5:47 The free template and the one-line install
6:45 Reviewing a downloaded skill before it goes in
8:44 Why this demo runs in Terminal
9:57 A fresh account, installed from zero
12:37 Plugin commands and auto-update
13:42 Adding the marketplace in the browser
16:23 Installed, plus the friction it caught
17:57 A house voice that updates itself
19:13 A rule becomes a check, Vale and em dashes
20:04 Teams, a plugin per department
21:02 Count the places your skills live

#ClaudeCode #AIAgents #AIAutomation #GitHub #AIForBusiness

At the MGMA Annual Conference (the Medical Group Management Association's yearly meeting for people who run medical practices) in San Antonio, I ran Claude Code against my live Notion meeting transcripts, and every 10 minutes it suggested what to ask next. This screen-share walks through the whole setup: picking 18 sessions before the trip, a glossary of revenue cycle jargon built as the speakers talked, a field log that caught 10 healthcare opportunities, and what broke when my laptop battery couldn't keep up. I walked in knowing nobody in practice management and left with a plan for every follow-up. Take any piece of it for your next event.

AI for medical groups and practices:
https://seriousai.partners/healthcare/?utm_source=podcast&utm_medium=description&utm_content=ep-328-ai-conference-questions

Serious AI News, every Monday. AI for operators, what we built and what it cost:
https://seriousai.partners/newsletter/?utm_source=podcast&utm_medium=description&utm_content=ep-328-ai-conference-questions

Tell us what's breaking:
https://seriousai.partners/contact/?utm_source=podcast&utm_medium=description&utm_content=ep-328-ai-conference-questions

Chapters:
0:00 Every 10 minutes, my MacBook told me what to ask
1:46 Why a healthcare conference
2:34 Picking 18 sessions with Claude before the trip
5:37 The first session, a fire hose of jargon
7:33 Long Claude Code threads hold up now
10:21 Live questions every 10 minutes
11:38 Why I never asked a question
13:33 A running glossary, built live
14:57 Composio and more than one Google account
17:14 What Claude can't see in the room
18:27 The routine Claude built for itself
20:10 The battery problem and the phone workaround
23:43 Slides in, field log out
24:21 Lessons, fewer sessions and more meetings
25:35 Was I the agent?
28:49 Why I'd never skip this setup again

#ClaudeCode #HealthcareAI #AIForBusiness #AIAgents #ConferenceStrategy

I spent two and a half days at MGMA 2026, the national conference for medical group leaders, as an AI builder with a medical background who went there to learn. What I found is a blind-spot triangle: are you paid what your contracts say, is your AI's work right, and does anyone own the whole picture across the tools you bought one at a time. Payers already use AI to deny claims and quietly downcode them, and many of the practices in those rooms are still answering by hand. Take one question back this week: for an AI tool you pay for, what single number would prove it's working?

AI for medical groups and practices:
https://seriousai.partners/healthcare/?utm_source=podcast&utm_medium=description&utm_content=ep-327-underpaid-no-denial

Serious AI News, every Monday. AI for operators, what we built and what it cost:
https://seriousai.partners/newsletter/?utm_source=podcast&utm_medium=description&utm_content=ep-327-underpaid-no-denial

Tell us what's breaking:
https://seriousai.partners/contact/?utm_source=podcast&utm_medium=description&utm_content=ep-327-underpaid-no-denial

Chapters:
0:00 Buying AI one tool at a time
0:16 MGMA 2026, going in to learn
2:50 AI in every room, and the pressure on operators
4:21 Point tools and ambient scribes
5:36 Only a few run a mature AI program
6:57 Payers already automate denials
8:16 Why the buying decisions are happening now
9:50 The blind-spot triangle
12:26 In-house build versus vendor tools
14:33 Silent downcoding, billed one code and paid another
15:45 Contracts that pay less than they say
17:09 A scribe that ignored its rules
18:10 Coders still checking by hand
19:11 The human side of automation
20:20 Who owns the whole picture
21:51 Your move, name one number

#HealthcareAI #MedicalPractice #RevenueCycle #AIImplementation #EnterpriseAI