Governed Vibecoding vs Unmanaged AI CodingRead Now →
Skip to main content

Claude Skills

Learn how Claude Skills use SKILL.md, project and personal folders, supporting files, invocation rules, and governance controls in Claude Code.

12 min read
Axiom Studio Team· Engineering

How Claude Skills Work

Claude Skills turn repeated procedures into discoverable Claude Code capabilities. They are strongest when the SKILL.md stays short, names the trigger clearly, and points to supporting files only when the task needs them.

Source checked against Claude Code skills docs

Official documentation source

Claude Code skills are SKILL.md files that extend Claude with reusable instructions.

Skills can be invoked directly with slash commands or loaded automatically when relevant.

Personal, project, enterprise, and plugin scopes have different precedence and sharing behavior.

Primary file

SKILL.md

Common project path

.claude/skills/<skill-name>/SKILL.md

Invocation

Automatic relevance match or /skill-name

Can bundle

References, examples, templates, scripts

Good fit

Repeatable coding, review, docs, analysis, deployment, or verification workflows that teams keep prompting by hand.

Watch closely

Skills with allowed tools, shell injection, dynamic context commands, or references that can steer high-privilege changes.

Governance move

Treat project skills like code: code review them, version them, test trigger phrases, and document owners.

Claude Skills are reusable instruction packages for Claude Code. A skill teaches Claude when a workflow applies and what procedure to follow when it is active.

The key architectural point is progressive disclosure. Claude can see enough metadata to choose a skill, then load the detailed SKILL.md only when the task calls for it. That keeps everyday context smaller than a giant always-on CLAUDE.md file.

Where Claude Skills Live

Claude Code guide

Claude documents personal, project, enterprise, and plugin skill scopes. Project skills commonly live at .claude/skills/<skill-name>/SKILL.md and can be committed with the repository.

Claude also treats command files and skills as closely related. Existing .claude/commands files continue to work, while skills add richer folders, supporting resources, and invocation controls.

1

Personal skills apply across projects for one user.

2

Project skills apply to a repository and can be shared by the team.

3

Plugin skills are distributed with a plugin namespace.

4

Enterprise or managed skills can provide organization-wide guidance.

Discovery and Invocation

Claude Code guide

Claude can load a skill because the user's prompt explicitly invokes it or because the skill description matches the current task. The description is routing metadata, so it should be specific and front-load the use case.

Project skills are discovered from .claude/skills directories along the relevant project path. Claude Code also watches skill directories for SKILL.md changes during a session, with some plugin-related changes requiring plugin reloads.

Do not over-port platform behavior

Each platform can use SKILL.md-style files, but discovery paths, invocation rules, frontmatter, permissions, and distribution mechanics differ. Verify the target platform before copying a skill unchanged.

What to Bundle

Claude Code guide

A Claude skill can include more than a single markdown file. Supporting examples, reference documents, templates, and scripts keep the main SKILL.md concise while still giving Claude reliable task-specific context.

The practical rule is to keep routing and the main procedure in SKILL.md, then reference heavier material by filename. If the model needs a spec, checklist, or helper script, tell it when to open or run that file.

Permissions and Security

Claude Code guide

Claude skill frontmatter can influence how a skill is invoked and which tools are available while it runs. That makes skill review a security activity, not just a documentation task.

Review allowed tools, command injection, scripts, and dynamic context carefully. A skill that pulls live diffs or runs shell helpers may be useful, but it also changes what the model sees and can do.

1

Require review for project skills before they land in the repository.

2

Use explicit stop conditions for production, secrets, and destructive actions.

3

Keep sensitive examples, tokens, and customer data out of skill folders.

4

Test whether the skill triggers too broadly before enabling implicit use.

Enterprise Governance Checklist

Claude Code guide

For enterprise Claude Code usage, the governance question is whether teams can explain which skill ran, what tool access it had, who approved it, and what output it produced.

Claude Skills fit best when paired with tracked work items, code review, audit logs, and a review cadence for skill changes.

Govern Claude Skills with tracked agent work

VibeFlow connects reusable agent workflows to work items, execution logs, commit records, security review, QA, and durable project context. That audit trail makes skills reviewable instead of invisible prompt behavior.

See VibeFlow

Ready to get started?

See how Axiom Studio can transform your AI infrastructure with enterprise-grade governance, security, and cost optimization.

Contact Us