Most businesses have processes that work. A way to onboard new clients. A format for weekly reports. A checklist for closing deals. The problem is these processes live in someone's head, or in a document no one reads, or get done differently every time.
Skills solve this.
What a skill is
A skill is a saved playbook your agent follows exactly. You write it once — step-by-step instructions, with context about what to check and how to format the output — and the agent executes it the same way every time you invoke it.
Unlike a workflow, a skill isn't triggered automatically. It's invoked by keyword in conversation ("Run the weekly report skill") or automatically by the agent when it recognizes a matching context.
Writing a good skill
The best skills are specific and action-oriented. Don't write "Summarize emails" — write:
1. Open Gmail 2. Filter to emails received in the last 24 hours, starred or from people in my contacts 3. Group by sender domain 4. For each email, write one line: who it's from, what they want, whether it needs a response today 5. Flag anything mentioning "urgent", "invoice", or a deadline within 3 days
Specificity removes ambiguity. The agent follows the list; it doesn't interpret it.
Trigger keywords
Add trigger keywords to a skill and the agent will invoke it automatically when those words appear in conversation or in a prompt. A skill called "Inbox triage" with keywords ["triage", "emails", "inbox"] will fire whenever you say "Can you triage my inbox from this morning?"
Skills as institutional memory
The long-term value of skills is encoding your best practices. When you discover a better way to run a weekly review, update the skill. Every agent, every team member, and every future session benefits.
Think of it as teaching the system, not just completing a task.