Episode 4: Skill - Reusable Workflow Templates

⏱ Est. reading time: 4 min Updated on 5/7/2026

A Skill is a "recipe" defined in Markdown. It doesn't contain executable code itself, but rather uses instructions to tell Claude which steps to take and which Tools to call in specific situations.

Anatomy of a Skill

A Skill is centered around a SKILL.md file, which typically includes:

  • name: The name of the Skill (registered as a /name slash command).
  • description: A description of when this Skill should be triggered.
  • allowed-tools: A whitelist of Tools this Skill is permitted to use.
  • Instruction Body: Detailed execution logic (Objective, Instructions, Constraints).

Three Ways to Install

Method Directory Use Case Example
Plugin Auto-install ~/.claude/plugins/cache/<author>/<name>/skills/ Installed via the plugin system claude-mem skills
Global Manual ~/.claude/skills/<name>/SKILL.md Frequently used manual workflows GSD's 137 skills
Project-level Project/.claude/skills/<name>/SKILL.md Logic specific to one project Deployment scripts

GSD: A Masterpiece of Pure Skills

GSD (Get Shit Done) installs 137 Skills, almost entirely consisting of pure Markdown. It orchestrates built-in Tools like Read, Write, Edit, and Bash to implement highly complex engineering processes.

Why GSD doesn't need MCP:

  • Everything GSD does (reading/writing files, running tests, task decomposition) is within the native capabilities of Claude Code.
  • It uses highly structured Markdown instructions to give the AI a rigorous engineering mindset.

The Skill Lifecycle

stateDiagram-v2
    [*] --> Registered: Claude Code starts
    Registered --> Registered: Scans skills/ directory
Parses all SKILL.md files

    note right of Registered
        Skills appear as /name
        in the slash command list
    end note

    Registered --> Matched: User enters /gsd-quick
    Matched --> Injected: SKILL.md content injected into System Prompt
    Injected --> Executing: Claude follows instructions
    Executing --> CallingTools: Calls allowed-tools
    CallingTools --> Executing: Tool returns result and continues
    Executing --> Completed: Task finished
    Completed --> Registered: Back to ready state