Skills are reusable agent files that give Claude Code or Codex a specific role and set of instructions. Each skill has two files:
skill-name.md— The agent file. This is what Claude reads. It contains frontmatter (name, description, model) and the agent's instructions. You drop this into.claude/agents/in your project.description.md— A human-readable description. This explains what the skill does, how to use it, and what options it has. This is what you read when deciding whether to download it.
Browse the skills/ directory to see what is available.
- Find a skill in
skills/that fits your task. - Read the
description.mdto see what it does. - Copy the agent
.mdfile to.claude/agents/in your project. - Run it by name: "run the paper-reviewer agent."
Or just ask Claude to do it:
Go to https://github.com/Harvard-Agentic-Science/Skills and download the paper-reviewer skill. Save it to .claude/agents/paper-reviewer.md in my project.
Each skill needs both files: the agent file and the description. Fork the repo, add a folder under skills/ with both, and open a pull request. Use the templates in template/ as a starting point:
template/skill-template.md— Template for the agent filetemplate/description-template.md— Template for the description
See CONTRIBUTING.md for the full walkthrough.
For background on agentic workflows and how skills fit into a research workflow, visit https://ai.physics.harvard.edu