SkillProof tested Claude skills

Reference

The Claude Code skills glossary

26 terms you run into the moment you open your first SKILL.md — defined in plain English, checked against the official docs and against the skills we've actually installed. Where a term is easier to see than explain, we link a tested skill.

A

4 terms
  1. Agent Skill An Agent Skill is a folder containing a SKILL.md and supporting files, extending AI agent capabilities.
  2. AGENTS.md AGENTS.md is a convention for repository instructions for coding agents.
  3. allowed-tools SKILL.md frontmatter field granting Claude permission to use specific tools without asking.
  4. Auto memory Auto memory stores Claude's notes to itself across sessions in markdown files.

C

5 terms
  1. CLAUDE.md CLAUDE.md is a markdown file of persistent instructions Claude reads at the start of every session.
  2. .claude/rules/ The .claude/rules/ directory holds project-specific instructions, splitting CLAUDE.md into smaller topic files.
  3. Compaction Compaction summarizes conversations to free context window space, automatically or on command.
  4. Context window The context window is the total span of tokens the model can see at once, including prompts, definitions, history, and file contents.
  5. context: fork Runs a skill in isolation within a subagent, using only the skill's content as the prompt.

D

1 term
  1. disable-model-invocation A SKILL.md frontmatter field that prevents Claude from automatically loading a skill; only you can invoke it.

F

1 term
  1. Frontmatter Frontmatter is the YAML block at the top of a SKILL.md file that configures the skill's behavior and metadata.

H

1 term
  1. Hooks Hooks are user-defined commands or endpoints that run automatically at specific points in Claude Code's lifecycle.

M

3 terms
  1. MCP (Model Context Protocol) MCP (Model Context Protocol) is an open source standard enabling Claude Code to access external tools, databases, and APIs directly.
  2. MCP vs skill MCP servers grant Claude new capabilities by connecting to external systems; skills provide knowledge and procedures.
  3. Model invocation Model invocation is Claude loading a skill automatically when your request matches the skill's description.

P

4 terms
  1. Personal vs project skills Skills can live locally for personal use, within a project, or as part of a plugin.
  2. Plugin A plugin bundles Claude Code extensions like skills, agents, and servers into a distributable package.
  3. Plugin marketplace A catalog distributing plugins, offering discovery, version tracking, and updates from sources like git.
  4. Progressive disclosure Progressive disclosure loads skill content on demand, reducing initial context cost.

S

5 terms
  1. Skill description The description frontmatter field tells Claude Code when to apply a skill.
  2. Skill vs prompt A skill is a saved, reusable prompt; a prompt is a one-off instruction for a single turn.
  3. SKILL.md SKILL.md is a markdown file defining a Claude Code skill's behavior, comprising YAML frontmatter and markdown instructions.
  4. Slash command A slash command is what you type to invoke a skill directly, like /skill-name.
  5. Subagent A subagent is a specialized assistant with its own context, prompt, tools, and permissions for specific tasks.

U

1 term
  1. user-invocable A SKILL.md frontmatter field that controls whether a skill appears in the / menu.

W

1 term
  1. when_to_use when_to_use is an optional SKILL.md frontmatter field for Claude skill trigger phrases.