What exactly is a «skill» in Claude Code?

A skill is a Markdown file (.md) containing the instructions for a specific workflow. It's stored in .claude/commands/ and invoked with /skill-name. Claude reads the file and executes the workflow according to the defined instructions. Skills can be as simple as a formatting command or as complex as a multi-step pipeline.

How do I know if a skill is working well?

With evaluations (evals). An eval is a controlled test: you run the skill on a known input and check if the output meets the defined criteria. The Skill Creator tool includes functionality to run benchmarks and measure output variance. A skill without an eval is an unverified promise.

Global skills vs. project skills — which do I use when?

Global skills (in ~/.claude/commands/) are available in all projects. Project skills (in .claude/commands/ inside the repo) only apply to the project where they're defined. The recommendation: start with project skills. Only promote to global when a skill is genuinely universal and doesn't create naming conflicts with other projects.

Can a skill call another skill?

Yes — this is called a higher-order or composite skill. For example, a /deploy skill could invoke /test, /build, and /push in sequence. This composability is one of the most powerful features of Level 5: you can build complex pipelines from simple, tested building blocks.

Level 5 — The Skilled

13 March 2026

Mindset: «Claude becomes YOURS»

At Level 5, you stop teaching Claude how to do things in every session. Your configuration already knows. You've encoded your standards, your processes, your way of working — and now Claude executes them consistently, without you having to repeat it.

A skill is a workflow you've articulated so well that Claude can execute it reliably with a single command. At this level, your setup is different from anyone else's — and that difference is a real competitive advantage.

You're here when...

You create custom skills that encode YOUR standards and workflows
Your Claude setup is unique — nobody else's works like yours
You use /skill-name to invoke workflows instantly
Skills reference other skills for composable pipelines
You run evaluations to confirm a skill actually works

Skills to master

Skill authoring — articulate your workflow so Claude can execute it consistently
Slash commands (.claude/commands/) for instant, reliable invocation
Workflow skills: multi-step pipelines, higher-order skills (skills of skills)
Skill Creator tool — build, test, benchmark, and optimise your own skills
Project-scoped skills to avoid global trigger conflicts
Running evals and benchmarks to confirm a skill actually works

Trap: Skill overload

Too many skills = Claude can't pick the right one. 20–30 curated, project-scoped skills beats 1,000 generic ones. The more options in the global list, the lower the hit rate.

The antidote: keep skills local to the projects where they make sense. Use /skill-name for guaranteed invocation when global scope applies. And run periodic evaluations — don't assume a skill that worked last week still works the same way today.

How to unlock Level 6

The unlock is recognising that one agent isn't enough. There are tasks that could run in parallel — auth and frontend and payments simultaneously, rather than sequentially. For that you need multiple Claude instances working at the same time. That's Level 6.

Next steps

Custom skill development for your team is one of the most valued services in the AI-IDE Mastery Programme. We work with you to articulate, build, and optimise your most critical workflows.

Frequently asked questions

What exactly is a «skill» in Claude Code?: A skill is a Markdown file (.md) containing the instructions for a specific workflow. It's stored in .claude/commands/ and invoked with /skill-name. Claude reads the file and executes the workflow according to the defined instructions. Skills can be as simple as a formatting command or as complex as a multi-step pipeline.
How do I know if a skill is working well?: With evaluations (evals). An eval is a controlled test: you run the skill on a known input and check if the output meets the defined criteria. The Skill Creator tool includes functionality to run benchmarks and measure output variance. A skill without an eval is an unverified promise.
Global skills vs. project skills — which do I use when?: Global skills (in ~/.claude/commands/) are available in all projects. Project skills (in .claude/commands/ inside the repo) only apply to the project where they're defined. The recommendation: start with project skills. Only promote to global when a skill is genuinely universal and doesn't create naming conflicts with other projects.
Can a skill call another skill?: Yes — this is called a higher-order or composite skill. For example, a /deploy skill could invoke /test, /build, and /push in sequence. This composability is one of the most powerful features of Level 5: you can build complex pipelines from simple, tested building blocks.