CHAPTERS
- 0:03 – 0:34
Why repeating instructions wastes time—and what “skills” solve
The video opens by highlighting the repeated friction of re-explaining coding standards, PR review expectations, and commit message formats. It introduces “skills” as the mechanism to teach Claude something once and have it applied automatically when relevant.
- •Common repetition: coding standards, PR feedback structure, commit formats
- •Skills eliminate re-teaching by making guidance reusable
- •A skill is authored once and reused automatically
- •Motivation: consistency and time savings
- 0:34 – 0:49
What a skill is: Markdown-based guidance + agent resources
This section defines skills concretely. Skills can be a single Markdown file (in Claude Code) or broader “agent skills” as folders containing instructions, scripts, and resources discoverable by agents.
- •Skill (Claude Code): an MD file that teaches Claude how to do a task
- •Agent skills can be folders with instructions/scripts/resources
- •Goal: improve accuracy and efficiency through reusable knowledge
- •Designed to be discoverable and reusable
- 0:49 – 1:05
How Claude decides to use a skill: description matching and activation
Claude selects skills by comparing your request to the descriptions of available skills. When there’s a match (e.g., asking for a PR review), Claude activates and reads the relevant skill to guide its behavior.
- •The skill description drives discoverability and selection
- •Claude matches your request against skill descriptions
- •Matching skills are activated automatically
- •Example: PR review request triggers PR review skill
- 1:05 – 1:20
Where to store skills: personal vs. project (team) scope
The video explains two main storage locations depending on who should benefit. Personal skills live in your home directory and apply across projects; project skills live in the repo and travel with the codebase for anyone who clones it.
- •Personal skills: ~/.claude/skills, follow you across projects
- •Use personal skills for preferences (explanations, doc formats, commit style)
- •Project skills: repo_root/.claude/skills, shared with the team
- •Use project skills for team standards and shared conventions
- 1:20 – 1:35
Team standards and brand guidelines as skills
Beyond coding workflow, skills can encode organizational standards such as brand guidelines used in web design. This section emphasizes using skills as the canonical place for shared, repeatable guidance.
- •Skills can encode company/team standards
- •Examples: brand guidelines, fonts, colors for web design
- •Shared skills ensure consistent outputs across contributors
- •Keeps institutional knowledge close to the work
- 1:35 – 2:06
Skills vs. other customization: Claude.md, on-demand loading, and context efficiency
This chapter contrasts skills with other Claude Code customization mechanisms. Claude.md loads into every conversation, while skills load only when relevant—reducing context bloat and keeping task-specific checklists out of unrelated work.
- •Claude.md applies globally to every conversation
- •Use Claude.md for always-on preferences (e.g., TypeScript strict mode)
- •Skills are task-specific and load on demand
- •Only name/description are considered until a match, saving context
- 2:06
Skills vs. slash commands: automatic behavior without manual triggering
Unlike slash commands that require explicit invocation, skills are applied automatically when Claude recognizes the situation. This makes them more seamless for recurring workflows.
- •Slash commands require you to type/trigger them
- •Skills apply automatically when the scenario matches
- •Reduces friction and improves consistency
- •Better for recurring standards than manual commands
Best uses and the rule of thumb: if you repeat it, write a skill
The closing emphasizes where skills provide the most value: specialized, repeatable knowledge tied to specific tasks. It ends with a simple heuristic—anything you find yourself repeatedly explaining is a candidate skill.
- •Best for specialized, task-specific knowledge
- •Examples: code review standards, commit message formats, brand guidelines
- •Improves reliability by standardizing recurring outputs
- •Rule of thumb: repeated instruction = skill waiting to be written
