October 7, 2026
[https://docs.google.com/presentation/d/1ShTfJUnr-WS-YDoPFoCJbHWaX-Jh_zzHL_P5FiBTkSc/edit?usp=sharing](https://docs.google.com/presentation/d/1ShTfJUnr-WS-YDoPFoCJbHWaX-Jh_zzHL_P5FiBTkSc/preview?usp=sharing)
A harness that finds agent skills on the web, collects them into Notion, tags them, and holds them for your review. Claude does the work through the Notion connector. Notion holds the state.
You end up with two Notion databases and four repeatable prompts:
| Stage | What Claude does | Reads | Writes |
|---|---|---|---|
| Explore | Finds places where skills are published | The web | Sources, Status = Proposed |
| Seek | Pulls skills from each source | Sources where Status = Active | Skills, Status = New |
| Organize | Tags, summarizes, flags duplicates | Skills where Status = New | Skills, Status = Organized |
| Review | Nothing. You decide. | Skills where Status = Organized | Approved or Rejected |
| Understand | Answers questions and generates from the collection | Skills where Status = Approved | A digest page, a comparison, or a new skill |
The point of the exercise: a Git repo or an Obsidian vault can store the skill files, but the Status column is what turns storage into a queue. Each stage picks up rows in one status and leaves them in the next.
The same harness packaged as a reusable skill is in the two sub-pages at the bottom of this page.
You need three things, and about 45 minutes.
How to use this guide:
This proves Claude can read and write your workspace, and gives the harness one place to live.