⊕ Advanced

Skills Engineering: Design, Test, and Share Agent Skills

A cross-platform advanced track on writing, testing, versioning, and distributing agent skills for Claude Code and Hermes — treating skills as production software, not throwaway text files.

📚 13 lessons 🕐 187 min total ✓ Free

Skills as Software

Understand why skills accrue technical debt, how the Level 0 scan makes every unused skill a recurring cost, and the five-stage lifecycle that keeps a library healthy.

The agentskills.io Standard: One Format, Every Platform

Learn the open agentskills.io format — which six fields travel unchanged across Claude Code, Hermes, Cursor, and dozens of other agents — and when to write for portability versus platform power.

Skill Design Principles

Apply single responsibility, reference versus task framing, composition patterns, and advanced frontmatter constraints to design skills that stay useful as your library grows.

Write, Capture, or Install

Choose the right authoring path for each skill — hand-writing for full control, agent capture for completed workflows, or community install — and understand the quality tradeoffs each path carries.

Hooks in Skills

Register session-wide automated behaviors the moment a skill is invoked — hooks persist for the entire session and turn skills into persistent configuration tools, not just instruction documents.

Testing Your Skills

Verify that a skill fires on the right prompts and measurably improves output using baseline A/B sessions, the skill-creator eval plugin for Claude Code, and Hermes CLI testing with skill_view.

Measuring Effectiveness and Tuning Auto-Invocation

Track invocation frequency and context cost with /skill-doctor and Hermes Curator, understand auto-compaction token budgets, and tune when_to_use and paths for precise auto-invocation.

Versioning and Maintaining Skills

Treat skills like code — use git for history, semantic versioning for shared skills, deprecation patterns for safe retirement, and a review schedule to catch drift before it reaches production.

Building Your Personal Library

Design a personal skills library that scales — naming conventions, location strategy, the experiment-to-stable lifecycle, a regular audit schedule, and the one directory name you must never use.

Sharing Skills on GitHub

Structure a public skills repository, declare compatibility and license, contribute to the Skills Hub, and write skills that install cleanly with no hidden dependencies.

Team and Enterprise Distribution

Distribute skills across a team or organization — managed deployment, project-level version control, skillOverrides governance, and the cloud sync gotcha that silently drops personal skills in Cowork sessions.

The Community Ecosystem

Navigate the 90,700-skill community ecosystem — understand trust tiers, apply the seven-step evaluation checklist, and learn what the February 2026 Koi Security audit revealed about real-world community skill risk.

Capstone: Build Your Skill Engineering Workflow

Apply every concept from this track in one end-to-end walkthrough — build, test, version, and publish a production-quality review-pr skill from idea to public distribution.

Start Lesson 1 →