Author(s): Divy Yadav

Originally published on Towards AI.

After testing more than 100 Claude Code skills, only six became part of my daily workflow. Here is what each one does.

I built more than 100 Claude Code skills.

The author explains that most of the skills they built weren’t truly helpful because they solved hypothetical problems rather than day-to-day ones; the six that stuck share a consistent design pattern: each skill has a narrow scope (does one job), injects dynamic real-time context into Claude before it answers (using the ! command syntax like !git diff HEAD), and includes explicit stop conditions so Claude won’t continue into unwanted steps. They then walk through what these skills look like in practice—covering changes summarization, instant Conventional Commits-style commit generation, a security-focused review that restricts feedback to security findings, a GitHub-issue-driven fixer that uses $ARGUMENTS for parameterized issue fetching, a docs-checking tool that pulls up-to-date documentation via llms.txt (instead of relying on model training cutoffs), and a guarded “ship” workflow that runs tests and deploy steps only when invoked directly. Finally, they advise how to build such skills first (don’t jump straight to /ship), and highlight a mistake they made early on: skills become less effective when they are too long, so instructions should be concise and action-focused.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI