๐—ฆ๐˜๐—ผ๐—ฝ ๐—ฅ๐—ฒ๐—ฏ๐˜‚๐—ถ๐—น๐—ฑ๐—ถ๐—ป๐—ด ๐—ง๐—ต๐—ฒ ๐—ฆ๐—ฎ๐—บ๐—ฒ ๐—”๐—œ ๐—ฆ๐—ธ๐—ถ๐—น๐—น๐˜€

Your team is wasting time.

One person builds an AI coding skill to run tests. Two weeks later, someone else builds a different version for the same task. They did not know the first one existed.

This creates chaos. You get different results on different machines. You get inconsistent standards. You get entropy instead of progress.

The solution is not more discipline. The solution is a shared catalog.

A good catalog must be easy to use. If adding a skill feels like a chore, your team will not do it. The catalog will die.

I use a single repo for my catalog. It follows a simple structure:

At the root, a master table acts as the single source of truth. If a skill is not in the table, it does not exist.

To keep the catalog alive, I turned the bookkeeping into a skill. I call it "catalog-a-skill."

Instead of manual updates, the agent does the boring work:

The agent handles the tedious parts. The human does the fun part: writing useful skills.

To make this work for your team, follow this pattern:

  1. Create a catalog repo with simple categories.
  2. Set a clear source of truth, like a master table.
  3. Write a meta-skill to handle the updates.
  4. Add gates. The agent should refuse to add a skill if data is missing or names do not match.

A catalog survives only when contributing costs nothing. Let your agent keep the books so your team can keep building.

Source: https://dev.to/edpittol/stop-your-team-from-rebuilding-the-same-ai-skills-a-shared-catalog-that-maintains-itself-35l5

Optional learning community: https://t.me/GyaanSetuAi