Descriptions:
Anthropic’s official skill creator plugin for Claude Code gets a thorough walkthrough from Nate Herk, who first lays out a conceptual framework for understanding skills before live-building a new one from scratch. The central distinction he introduces — capability uplift skills versus encoded preference skills — is the most useful organizing idea in the video. Capability uplift skills improve Claude’s output on tasks it could already attempt (like front-end design), while encoded preference skills lock in specific sequential workflows unique to the user, making them more durable across model upgrades.
The skill creator plugin, installable via the /plugins menu in Claude Code, adds four capabilities: creating new skills from natural language descriptions, modifying existing ones, running evaluations to measure performance, and tuning trigger phrases so the agent reliably calls the right skill. Herk walks through a trigger tuning evaluation graph sourced from Anthropic showing accuracy improvements before and after optimization, with green and blue bars illustrating train and test score gains.
The live demo shows Herk building a “YouTube weekly roundup” skill inside his personal Herk 2 project in VS Code — starting from a plain-language description and watching the skill creator generate, test, and refine the skill structure. He closes with a quote from Anthropic suggesting that future models may need only a high-level natural language spec to generate complete, working skills, pointing toward a near-term future where skill authoring requires far less manual prompt engineering.
📺 Source: Nate Herk | AI Automation · Published March 05, 2026
🏷️ Format: Tutorial Demo
