Claude Code Skills Are Broken (Beginner to Pro)

Detailed Summary

Intro & Live Demo (0:00 - 02:41)

The presentation begins with a demonstration of four AI agents running in parallel to handle daily planning, business pulse checks, diagram creation, and YouTube comment analysis.

• Parallel processing allows for massive productivity gains without context switching.
• The 'Morning Coffee' skill automates calendar management and task prioritization by connecting to ClickUp.
• Visualizations like Excalidraw diagrams can be generated with perfect text and layout through dedicated skills.

What are Skills? (02:41 - 05:24)

Skills are defined as reusable instructions saved as markdown files that ensure consistent AI output.

• They function as SOPs (Standard Operating Procedures) for AI agents.
• Benefits include personal productivity, team leverage (turning existing SOPs into automations), and potential monetization.
• Skills can run scripts, call APIs, and even delegate tasks to sub-agents.

Anatomy of a Skill & Reference Files (05:24 - 11:00)

A deep dive into the technical structure of a skill file and how it manages data.

• Structure: Skills live in the .claude/skills/ directory. The skill.md file contains YAML front matter (name/description) and the workflow steps.
• Data Management: There are two options for organizing files: self-contained (all scripts/references in the skill folder) or project-wide (pointing to paths elsewhere in the project).
• Efficiency: Using markdown reference files is faster and cheaper than making constant API calls or web searches.

How Claude Triggers Skills (11:00 - 13:17)

Claude uses a three-level progressive loading system to manage context and tokens.

• Level 1: Claude searches only the name and description (approx. 100 tokens).
• Level 2: Once a skill is selected, it loads the full skill.md (1,000+ tokens).
• Level 3: Extra files (scripts/templates) are only loaded if the specific task requires them.
• Skills can be triggered via explicit slash commands (e.g., /morning-coffee) or natural language detection.

The Feedback Cycle & Building Framework (13:17 - 18:10)

Creating effective skills is an evolutionary process based on observation.

• Optimization Example: Hardcoding ClickUp list IDs into a skill saved tokens by preventing the agent from searching for IDs every time.
• Six-Step Framework:
  1. Define Name and Trigger.
  2. Set a clear Goal.
  3. Outline the Step-by-Step process.
  4. Identify necessary Reference Files.
  5. Establish Rules and guardrails.
  6. Implement a Self-improvement loop.

Live Skill Build & Testing (18:10 - 23:55)

A step-by-step demonstration of building an 'Infographic Builder' skill.

• The process involves using a 'Skill Builder' agent to interview the user about the desired workflow.
• The agent automatically creates the folder structure, markdown files, and API integration scripts.
• Initial testing revealed a need for better logo transparency and aspect ratio constraints, which were fixed by updating the skill's rules.

Testing, Debugging & Global Skills (23:55 - 27:17)

Final tips for mastering skill deployment and troubleshooting.

• Common Fixes: If steps are wrong, edit instructions; if context is missing, add reference files; if the skill triggers too often, adjust the YAML configuration.
• Global vs. Local: Local skills stay within a specific project folder, while Global skills (stored in the home directory) are accessible across every project on the computer.
• Global skills are ideal for company-wide context, personal tone of voice, or universal design standards.

Notable Quotes

"I have genuinely never been as productive as I am right now because of Claude's skills... it just comes down to one simple word and that is leverage."

"Skills are basically SOPs for your AI agents. The same way where you would train a human employee by letting them read through an SOP... you just train an agent on it."

"Processing markdown files for your agent is so much quicker and cheaper than actually making API calls or HTTP requests."