In this section I review one AI-powered application and demonstrate how it can be used to create new value.
In Section 1 of this issue, I described how I use the Six Thinking Hats for innovation workshops. What I didn't mention is that I've since encoded that framework into an AI skill that anyone can install and use.
That's the story of this section. Skills, the open standard for teaching AI how you think, how you write, and how you create, are expanding far beyond coding. Three open-source projects show what's possible when you package expertise as portable AI instructions.
A quick refresher: what are skills?
We first covered skills in issue #5, when they had just come out and were a Claude-only feature. A lot has changed. In December 2025, Anthropic published the SKILL.md specification as an open standard. As of early 2026, over 30 AI agents support it: Claude Code, GitHub Copilot, Cursor, Windsurf, Gemini CLI, OpenAI Codex, and more. The ecosystem has grown to over 740,000 skills across marketplaces like SkillsMP and skills.sh.
The key distinction from other AI configuration mechanisms is that skills use progressive disclosure. When it starts up, your AI agent only loads the name and description of each skill. When a task matches, the full instructions are loaded on demand. This means you can have hundreds of skills installed without bloating your AI's context. It's efficient, portable, and, because it's an open standard, not locked to any single vendor.
Spotlight 1: encoding a thinking framework with Critical Validation
I built the Innovation skills repository as an open-source collection of 16 innovation skills spanning five phases, from Customer Discovery through Validation. We covered the earlier version in issue #11. The skill I want to highlight today is critical-validation, which implements the Six Thinking Hats as a structured AI workflow.
The sequence of hats matters: Red → White → Yellow → Black → Green → Blue → Red. You start and end with the Red Hat, gut reactions with no justification required. In between, you move through facts (White), benefits (Yellow), risks (Black), creative alternatives (Green), and summary and decision-making (Blue).
One design decision I'm particularly intentional about is that the Red Hat has strict enforcement. If the AI tries to rationalize an emotional response with a "because" statement, the skill rejects it. De Bono was clear that the Red Hat captures pure emotional signal. The moment you add justification, you've switched to a different hat. Encoding that discipline into a skill means the AI maintains it even when a human wouldn't.
Spotlight 2: teaching AI your voice with Writing Style Skills
If you've ever had AI write something "in your style" and gotten back something that sounded nothing like you, writing style skills solve that problem.
I use this type of skill for my own work, including this newsletter. One good open-source example works like this: the AI writes a draft, you edit it, and scripts analyze the differences to extract writing rules. Rules are categorized by confidence. High-confidence rules are automatically applied, lower-confidence ones await confirmation.
The clever part of this specific implementation is contrast-based learning. One edit (AI's original vs. your final version) carries twice the evidence weight of a direct writing sample, because the AI can see exactly what you changed and why. Each editing session makes the style profile more accurate, with fewer corrections needed over time.
In a way, a writing style skill is a Personalization flywheel in miniature. It collects your edits, uses them to produce better output, which means fewer corrections, which means faster writing. The more you use it, the more it sounds like you.
Spotlight 3: a complete creative toolkit with Story Skills
This is an end-to-end fiction writing toolkit built as five interconnected skills: story initialization, character management, worldbuilding, plot structure, and chapter writing. Everything is in plain markdown files, with the same structured approach developers use for code, applied to fiction.
What makes it work is consistency. A central "story bible" is referenced by all skills. Characters have identifiers tracked in index files, and locations and systems are cross-referenced. When the chapter-writing skill drafts a new scene, it pulls context from characters, worldbuilding, and plot files to maintain consistency across the entire narrative. The skill also comes with plot frameworks that include Three-Act, Hero's Journey, Save the Cat, and Kishotenketsu. I have started to experiment with this skill recently, and it does feel like a writing superpower.
Skills 2.0 and the evaluation flywheel
Claude recently released Skills 2.0. The approach is to improve skills through an evaluation flywheel: you collect outcomes from real usage, identify where the skill fails or underperforms, and iterate on the instructions. It's the same data flywheel concept from this issue's framework section, applied to the skill itself. The skills you use get better the more you use them, as long as you're actively feeding results back in.
The following video walks through how Skills 2.0 work and how to build them with an evaluation loop:
Your action step
Pick one framework, process, or style decision you apply repeatedly and haven't yet encoded. It could be a decision framework you use in meetings, a review checklist you run on every piece of content, or a domain-specific tone you want AI to maintain. Write the SKILL.md for it, even as a rough draft. Install it in Claude Code, Cursor, or whichever agent you use. Then run it on a real task this week and see what breaks. The first version will be imperfect. That's fine. The flywheel starts when you start iterating.
If you'd like to explore how skills and agentic workflows could systematize your team's expertise, or want me to run a working session on AI skills for your leadership team, I'd love to help.