# tutorial
How to Use Claude Code Skills
Once Invoked is connected, skill discovery is automatic. Here is what actually happens at runtime.
prerequisite
Invoked must be added as an MCP server. Setup takes 2 minutes →
What Happens at Runtime
- 1
You start a task
Write a commit message, review a PR, draft a README — any task. Claude detects what you are doing from context.
- 2
Claude queries the registry
Claude calls invoke_search_skills automatically. Invoked returns the most relevant skills for your task — ranked by match quality.
- 3
Claude fetches the skill
Claude calls invoke_get_skill to retrieve the full SKILL.md content — the instructions, workflows, and domain expertise inside it.
- 4
Claude follows the skill
The skill's instructions guide Claude's behavior for that task, exactly like instructions in your CLAUDE.md. The skill is not permanently installed.
Example — Commit Message Task
# you ask Claude to write a commit message
→ invoke_search_skills("write commit message for staged changes")
# Invoked returns: commit-commands:commit (relevance: 0.94)
→ invoke_get_skill("commit-commands:commit")
# Claude reads the SKILL.md — conventional commit format,
# scope rules, body guidelines
→ Claude writes the commit following the skill's instructions
Runtime Skills vs CLAUDE.md — When to Use Each
| Runtime skill | CLAUDE.md rule | |
|---|---|---|
| Applied | Per task, on demand | Every session, always |
| Scope | Any skill in the registry | What you write yourself |
| Maintenance | Registry updates automatically | You update manually |
| Best for | Broad task coverage | Team standards, mandatory rules |
Frequently Asked Questions
Do I need to tell Claude which skill to use?
No. Claude queries the Invoked registry automatically based on what you are working on. You do not need to name a skill or run a command — discovery is part of the runtime loop.
How does Claude know which skill is relevant?
Invoked matches your task context against skill descriptions in the registry using semantic search. The most relevant skill is returned first. If no skill is a strong match, the agent proceeds without one.
Is a skill applied permanently or just for one task?
Skills are applied per-task. The agent fetches the SKILL.md content at the start of a task and uses it as guidance. It is not installed into your agent permanently — the next task queries the registry fresh.
What if I want to use a specific skill every time?
Copy the skill's content into your CLAUDE.md or project rules file. That makes it a permanent instruction rather than a runtime-discovered one. Both approaches work — runtime discovery is better for broad task coverage, direct inclusion is better for mandatory team standards.
Can I see which skills Claude is using?
Yes. Claude Code shows all MCP tool calls in its output — you will see invoke_search_skills and invoke_get_skill calls with the skill name and the context it matched on. Skill creators see the same data in their Invoked dashboard.
11,000+ skills in the registry. Browse what's available.
Browse Skills →