Using skills
Finding a skill is different from loading it
Use discovery to choose a workflow, then confirm the full instructions reached the conversation.
Quick answer
Discovery gives the AI a short description of available skills. Loading retrieves the chosen skill’s instructions. In Skillset, list_my_skills is the library discovery tool and use_skill retrieves a specific workflow. Ask for the exact skill by title and confirm it was loaded before expecting the AI to follow it.
Why two steps are useful
A catalog description helps decide whether a workflow fits. It is not the workflow itself. Loading every available instruction set would also introduce methods unrelated to the user’s current job. The two-step pattern keeps selection separate from applying the selected method.
Know which list you are looking at
At the protocol level, MCP separates discovering available tools from calling a tool. Skillset adds a product-level distinction inside that interface: its discovery tool lists skills, while its loading tool returns the selected instructions. A client displaying “Skillset connected” has not necessarily done either.
Source: MCP tool specification
Use a request the AI can verify
Suppose you bought a customer-research skill and want to analyze interview notes. First ask for skills relevant to customer interviews. Pick the intended title, then ask the AI to load all parts and name it. Only after that should you supply the notes and ask it to apply the workflow.
Find my customer interview skill in Skillset. Load the full workflow and confirm its exact title. Then tell me what interview notes or research questions you need.Troubleshoot the smallest missing step
If no skills appear, check that the AI connection uses the same Skillset account as the purchase. If the title appears but loading fails, inspect the access or safety message returned by Skillset. If loading succeeds but the answer is generic, ask the AI to identify the relevant workflow step and show which supplied evidence it used.
Long instructions may arrive in several parts. Let the client retrieve the continuation before starting the task. Repeating a fragment in chat is not a substitute for loading the complete workflow; an omitted boundary or exception can materially change the intended behavior.
Common questions
Does “connected” mean “using my skill”?
No. Connection, discovery, loading, and applying the method are distinct steps. Check for tool activity and a named confirmation.
Do I need to type a tool name every time?
Usually plain language is enough when the AI app exposes the connection. Tool names are helpful when troubleshooting ambiguous behavior.
Can a public listing prove access?
No. Public descriptions help with discovery. Loading the protected instructions requires the connected account to have current access.
Put a workflow to work.
Connect your library to your AI, or turn your method into a skill.