PowerUp Coaching — Living Knowledge Base

2025-10-23_Mastermind

5 min read

2025-10-23 AI Mastermind for Leaders

Table of Contents

Session Overview

The October 23 session picked up where October 16 left off, with Lou demonstrating the AIM Writing Team skill in full production — a live walkthrough of the orchestrated pipeline in action, from a single prompt to a completed, edited, and publisher-packaged article. The session opened with the group catching up on what had been discussed before Lou arrived (his computer had gone “wonky” and he was working from a backup Mac Mini), then moved quickly into the Claude Skills deep dive.

The group’s prior week had been rich with sharing. Bally had encountered “Wendy,” an English coach living in Thailand who was earning recurring income from clients by delivering GPT-based avatar services — a grounding reminder that simpler implementations create real business value. Donald Kihenja had looked up what Claude had to say about its own Skills announcement, which generated a useful meta-conversation about whether the group was too fluent in AI to appreciate genuinely significant capability shifts.

The Skills demonstration was thorough: Lou showed how ChatGPT Atlas could be used to research the Claude Skills announcement, then walked through the full skill creation process — from a single prompt asking Claude to learn and then write a skill, through to the orchestrated writing team running in real-time, scoring each output, editing on the fly, and writing discoveries to a persistent memory JSON.

High-Signal Moments

  • The AIM Writing Team skill ran live: researcher → strategist → writer → editor → publisher, with each role scoring itself and the final output being a complete, packaged article including headlines, tone guidance, SEO tags, and target publication recommendations
  • The reflection loop embedded in the skill produced genuinely useful self-critique — the editor detected rhythm issues, weak opening hooks, and mechanical language, all of which were addressed in subsequent passes
  • Lou discovered in real-time that memory writes to the JSON file only happened during full orchestration, not when calling individual sub-roles — an important edge case to fix
  • The Flesh-Kincaid readability score on the first article came out at 11.2 (too high); Lou noted he targets 6–8 for his audience, and the rewrite toward Gary Vee/Seth Godin style brought it down appropriately
  • ChatGPT Atlas was assessed as a “basic remix of Chrome” — useful for AI-native browsing tasks but lacking the organizational features of standard browsers
  • The group probed Anthropic’s strategic motivation for Skills: Lou landed on “ecosystem lock-in through personalization” — once your expertise is packaged in a skill, switching costs rise substantially
  • Donald’s instinct to “just ask Claude what’s the big deal about Claude Skills” was a perfect demonstration of the group’s meta-AI fluency — using the tool to analyze the tool

Open Questions

  • How do we fix the JSON memory write to happen during sub-role calls, not just full orchestration?
  • Is the value of Claude Skills primarily in personal productivity, or in the emerging possibility of skills-as-IP that can be sold or shared?
  • What is the right Flesh-Kincaid target for different content types (newsletter vs. long-form article vs. LinkedIn vs. podcast)?
  • How does the writing team skill compare to human editors on genuinely high-stakes content — where are the remaining gaps?
  • What happens when two members of the group share the same skill — does the memory file create a personalization drift that makes it more useful over time?

Suggested Follow-Through

  • Fix the orchestrator skill to ensure memory writes happen even on sub-role calls by moving the “remember” instruction later in the prompt flow
  • Create at least one complete article using the full writing team skill pipeline, then compare it to a piece written without the skill — assess the quality difference honestly
  • Experiment with directly calling the Strategist sub-role for article architecture without running the full pipeline — useful for rapid content planning sessions
  • Document the skill installation process (create folder, write skill.md, zip, upload to Claude Capabilities) as a one-page quick-start guide for members
  • Share the skills zip file with at least one other mastermind member and observe how the same skill produces different outputs based on different prompts and contexts

Additional Resources

Links & Tools Shared in Chat

Books & Articles Mentioned

  • None

Ideas from Chat

  • Donald Kihenja asked Claude to evaluate one of his existing projects to determine whether it should be migrated to a Claude Skill — a useful meta-prompting workflow for AI architecture decisions — see Insight - Use Claude to Evaluate Whether a Project Should Become a Skill
  • Kasimir Hedstrom’s “board of 16” project: a customGPT/Gem/Claude project (not yet in n8n) — a personal advisory board of 16 expert personas the user can consult
  • Bally: “Gem is superb!” — Google Gemini Gems cited as a strong alternative to Claude Projects for persistent AI role configurations

Graph View