Build Your Own Skill
How to create Skills that automate your research workflows
Skills capture methodologies and expertise in a format AI agents can follow consistently. For example, how your lab reviews protocols, what a specific checklist covers, or how certain datasets should be formatted.
The easiest way to create Skills is through conversation.
Invoke the skill-creator Skill, describe your workflow naturally, and let the agent handle the formatting and structure. No technical expertise required.
Creating custom Skills with Claude
How to start
Start a new chat and describe the intended Skill:
- "I want to create a skill for reviewing IRB applications"
- "I need a skill that follows our lab's QC checklist for RNA-seq"
- "Help me create a skill for drafting grant specific aims"
Answer the AI agent's questions about the process. The goal is to provide enough detail that someone capable but unfamiliar could follow the approach. Specific answers produce more effective Skills.
Upload materials that demonstrate the approach (templates, checklists, examples, etc.). The AI agent uses these to understand the methodology.
Testing & Improving
How to test
Testing involves describing a task the Skill should address. "Using [skill name]" appears in the agent's thinking when the Skill activates. If something is off, the agent can update the Skill based on feedback.
How to improve
Skills improve through use. Starting simple and refining based on real tasks works better than trying to anticipate everything upfront.
Start with evaluation
Run the Skill on real-world scenarios and observe where the AI agent struggles or produces unexpected results. Pay attention to whether it triggers when expected.
Iterate and refine
When the AI agent goes off track, ask it to reflect on what went wrong and suggest updates. When it handles a task well, ask it to capture that approach.
What makes a good Skill?
Effective Skills typically contain:
When the workflow applies, what triggers it, what falls outside its scope.
The steps followed, what gets checked, what gets flagged, how output is structured.
Documents showing the preferred approach or format.
Policy guides, checklists, templates, playbooks, style guides, regulatory requirements.