Skip to guide

Building skills and workflows

Write down a method you already use

Start with a task you can explain and a result you can judge. State the goal, the inputs the agent needs, the steps it should follow and the checks that show it has finished.

A skill stores those instructions for reuse. In the file-based setup described in AI Tools & Skills, the entry point is a SKILL.md file. Scripts and reference material can sit beside it.

Make the result easy to check

Describe the output precisely. For a document review, that might mean a list of issues with quoted passages and proposed corrections. Include cases where the agent should stop, ask for missing information or request approval.

Try the method on a small real task. Inspect the output and record the corrections you make. Revise the instructions when the same misunderstanding recurs. Evals can turn those corrections into repeatable checks.

Connect steps when the task needs them

A workflow might gather evidence, analyse it, draft a result and check it. Give each step an observable output so that the next step receives the material it needs.

A plugin packages capabilities for installation. It may contain skills, tool connections and hooks that run at defined points. A schedule starts work at a chosen time. Keep the method, the package and the trigger clear when you describe the setup.

Skills and workflows are one part of the harness. They depend on the context and tools available to the agent.

Say hello

Draft a message about the work you need to complete.

Local draft preview. No message is sent. Review the draft, then copy it for use elsewhere.