Configure agent skills, tools, and credentials

Skills are on-demand instructions or knowledge the agent can consult. In the agent workspace, open Skills to assign an enabled skill or choose New skill. A skill includes a name, description, when-to-use guidance, and content.

Review skill content for accuracy and scope. Changes may require a new chat to take effect.

Tools are actions the agent can call. Assign only the tools required for the agent's role. When authoring or editing a custom tool, review its code, entry point, inputs, network destinations, and requested secrets before approval. Test with non-production data where possible.

Credentials store secrets such as API keys or tokens. Values are encrypted and write-only; after saving, you manage metadata rather than viewing the secret again. Give each credential a recognizable name.

Bind each secret requested by a tool to the intended stored credential. An unbound credential warning means the tool cannot use that secret correctly. Deleting a bound credential causes affected tools to stop working until they are rebound.

Start a new chat after material skill, tool, credential, prompt, or permission changes. Test a safe read-only request first, inspect the agent's result, and only then allow higher-impact actions.

Related articles: Configure an agent's handlers and permissions, Chat with an agent and review sessions, Understand API-key security and company scope.