Short answer: an Agent Skill is a folder with a SKILL.md file (instructions) and, optionally, scripts and reference files. Claude sees only the skill’s name and description until a task needs it, then loads the rest. It is the way to package your team’s know-how, like a runbook or a report format, so the assistant follows it every time.
Agent tooling changes fast. Commands and file formats below match the versions current at the time of writing; check the official docs for your version before relying on them.
Anatomy of a skill
structure
.claude/skills/postmortem-writer/
├── SKILL.md # instructions + front matter (required)
├── template.md # the document structure to fill in
└── scripts/
└── timeline.py # optional helper: builds a timeline from logsSKILL.md
---
name: postmortem-writer
description: Writes blameless incident postmortems in our company format.
Use when the user asks for a postmortem, incident report or RCA.
---
# Postmortem writer
1. Ask for: incident window, impact, timeline sources (logs, chat export).
2. If logs are provided, run scripts/timeline.py to build the timeline.
3. Fill in template.md exactly. Keep it blameless: describe systems and
decisions, never individuals.
4. Every action item needs an owner and a due date, or "MISSING".
5. End with "What went well", it is required in our format.Where skills work
| Product | How to add a skill |
|---|---|
| Claude Code | Folder in .claude/skills/ (project) or ~/.claude/skills/ (personal) |
| Claude apps (claude.ai, desktop) | Upload the skill as a zip in the capabilities settings (code execution must be enabled) |
| Claude API | Attach skills to requests through the API; see the Agent Skills docs |
Skills, commands, subagents, MCP: which one?
| You want to… | Use |
|---|---|
| Reuse a procedure or format the assistant should follow | Skill |
| Trigger a fixed prompt by name, like /review | Slash command (a saved prompt) |
| Delegate a task to a specialist with its own context and tools | Subagent |
| Give the assistant access to live data or external actions | MCP server |
| Run something automatically before or after a tool call | Hook (Claude Code) |
Skills can include scripts that run on your machine. Only install skills you wrote or reviewed, exactly like any other code.
And in ChatGPT?
There is no direct equivalent file format. The closest are Custom GPTs (instructions plus knowledge files plus actions) and Project instructions. They package know-how, but they are always loaded, so they compete for context instead of loading on demand.