0:00
/
0:00
/
Preview

Your Best AI Work Vanishes Every Session. 4 Prompts That Make It Permanent plus Access to My Skills Repo

Skills can be a powerful hidden context layer, but only if your agents can read them.

On March 11th, Anthropic slipped Skills into the sidebars of Excel and PowerPoint. The internet shrugged.

Which is fair. Skills have always had a Toyota problem: the workhorse nobody glorifies but that runs the world. The Camry doesn’t trend. It just keeps going.

Most people still think of skills the way they worked in October — as a personal prompting shortcut. You encoded your methodology into a Markdown file, stopped re-explaining it on every task, saved yourself some setup time. That framing was right when you were the only one calling your skill, sitting at a keyboard, with enough judgment to catch the output when it drifted.

Four things changed in five months, and together they break that framing completely. Agents now invoke skills as often as humans do, with no one watching and no one to redirect. Team admins provision skills across entire organizations with a single upload. The format became a cross-industry standard adopted by OpenAI, Microsoft, GitHub, and Cursor, with 500,000 skills now running across platforms interchangeably. And the same file that lived in a developer terminal now runs in Excel, PowerPoint, and M365.

Almost everything built to the October standard is underspecified for the March reality. The conventional approach, building a skill that works when you’re watching, quietly fails the moment the human leaves the loop. And the human is leaving the loop faster than most people have registered.

Here’s what’s inside:

  • The architecture and what people are building on it. Progressive disclosure, the specialist stack, the orchestrator pattern, and why the same file now runs from a developer terminal to an Excel sidebar.

  • How to build a skill that works in March. The description field where most skills fail, the five things every skill body needs, and why building from your outputs beats building from your intentions.

  • The failure asymmetry that changes everything. Why “good enough for my use” and “good enough for agents” are categorically different standards, and the four redesigns that close the gap.

  • The team-level case and the ecosystem gap. Three tiers most organizations get backwards, and why 500,000 skills exist and almost none are for knowledge work.

  • Four prompts to build your first skill this week. The backlog audit, the output-extraction builder, the agent-readiness stress test, and the team deployment planner.

Let’s get into it.

Subscribers get all posts like these!

User's avatar

Continue reading this post for free, courtesy of Nate.