Skills & Tools

Agents define roles

An agent establishes a specialized purpose and associated configuration. Choosing an agent does not grant additional permissions or make its answers authoritative.

Skills describe procedures

Skills provide structured guidance for approaching work. Their lifecycle and availability are governed rather than inferred from a request alone.

Tools expose actions

A tool provides a defined operation. Server policy and the user’s permissions determine whether an action can run; an instruction in conversation is not a substitute for authorization.

Review before relying

The Skills & Tools panel exposes governance information. Confirm that the intended capability is enabled and that its inputs, outputs, and boundaries fit the task.

Read the panel in context

Open the Dashboard and locate Skills & Tools where available. Resolved Skills describe guidance selected for the current context. Tools expose registered operations whose use is separately governed. Memory Notes are managed in the Memory panel. If an action is unavailable, request the appropriate operator review.

Ask Friday to guide you

In an authorized FridayLocalAI conversation, try:

Explain the Skills & Tools information I can see: [non-sensitive labels]. Distinguish reusable guidance, tools that can act, and permissions. For my task, identify what is available and what still needs operator confirmation.

Tell Friday what you can see and which step you have reached. If the instructions do not match your screen, say so; ask for the relevant Help topic or contact your operator. Friday cannot be assumed to see your screen or know your current settings.

Begin with the action you need

When a task sounds like something the assistant should “do,” separate the desired result from the mechanism. Explaining how to rename a file is different from renaming it. Drafting an email is different from sending it. Suggesting a sequence of steps is different from executing those steps against an actual system. This distinction helps you identify whether the task needs guidance, a Tool, or operator involvement. Describe the target and the expected outcome before asking for execution. If you only need an explanation, say so. If a real change is needed, confirm that the appropriate capability is configured and permitted. A confident response or an Agent’s specialized name does not establish that it can perform the operation in your environment.

What a Skill contributes

A Skill provides reusable guidance for approaching a kind of work. It can describe a procedure, expected checks, or a way to organize a result. The Skills & Tools panel can show resolved guidance for the current context where available. The application’s Skill handling is separate from the Memory Notes interface, even though both can influence how work is approached. Think of a Skill as an organized method rather than an unrestricted permission. A useful procedure might say how to review a source or prepare an artifact, but the actual operations still depend on available Tools and policy. If a step cannot run, the procedure does not authorize bypassing that limit. Ask what part can be completed and what requires a different approved capability or operator decision.

What a Tool contributes

A Tool exposes a defined operation through the runtime. Its inputs, target, permitted scope, and result matter. The existence of a registered Tool does not mean every user or Agent can call it for every purpose. Role, workspace, policy, and capability checks determine whether the requested operation can proceed. Before relying on a Tool result, understand what it proves. A lookup result can provide information; a file operation should produce a verifiable file state; an external action should have an appropriate confirmation or result. Do not stretch a narrow success into a claim that an entire workflow is complete. Ask Friday to separate the operation that ran from the remaining work and any uncertainty.

Read the panel in the selected context

Open Dashboard and inspect Skills & Tools where your role exposes it. Check that the intended project and conversation are active, because applicable guidance can depend on context. Read labels and status rather than assuming the panel is a shopping list of features that can all be enabled through a prompt. If something is unclear, describe the non-sensitive label to Friday and ask for an explanation. If a capability is missing, consult the operator about the intended configuration. Avoid making policy changes merely to get past an error. The useful outcome is a clear understanding of what the current task can do within the existing boundary and what additional review would be required.

Ask for a plan before a consequential operation

For a task involving changes, request a concise plan that identifies the target, the proposed action, and the result you will inspect. Include any boundary important to you, such as limiting work to a particular approved document or producing a draft without sending it. Clarify the point at which a separate decision is needed. After the operation, inspect the actual result. A plan and an attempted action are not completion evidence. If a result is partial, ask what succeeded and what remains unresolved. This is particularly important when a task mixes explanation, local data changes, and connected services. Treat each step according to what it does rather than assuming a single broad request resolves every authority question.

Use specialization without surrendering judgment

An Agent can help focus the assistant’s role, and Skills can make recurring work more consistent. Those benefits do not remove the need to review factual claims, source use, and completed actions. A specialized assistant can still misunderstand the task or lack an important capability in a particular deployment. Start with a small representative exercise and evaluate its result. For recurring work, keep a clear description of inputs, expected outputs, and review criteria. Use Memory Notes for appropriate durable preferences and maintain Skills through the governed process available to the operator. The combination is most useful when guidance, permission, execution, and review remain understandable as separate parts of the workflow rather than being hidden behind a promise of general autonomy.