Teach it your business logic.

A skill is an action the agent can take mid-conversation — check an order, book a slot, call your API — deciding for itself when to reach for it. A workflow is event-driven automation: it starts on a trigger like a new conversation or a label being added, then branches on conditions, waits with delays, and runs actions. Your installed skills show up inside workflows as actions too, so the two work together.

Mostly no. Install a Skill Set from the marketplace, drop in your keys, and enable it — no editor. Build a workflow from a template and wire it visually, and hit any REST endpoint with the Simple API Request node without a line of code. Code is the escape hatch: skills are JavaScript handler functions when you need custom logic the marketplace doesn't cover.

Each skill is a handler function that receives the configuration and secrets you set at install, plus the parameters the agent extracted from the conversation. It calls your API, and returns a result the agent speaks back. Sensitive values arrive separately in input.secrets, never mixed into ordinary config. Workflows can also call any endpoint directly with the Simple API Request node.

Skills run in an isolated sandbox with strict CPU and memory limits, and blocked APIs like eval, require, and process are rejected before a skill can deploy. Secrets you enter are stored on Cloodot and passed to the skill code at runtime, so before installing a third-party Skill Set: review the author and look for the verified badge, use scoped credentials with the minimum permissions, and rotate or revoke them when you uninstall.

A new conversation, an AI or human agent being assigned, a conversation opened, resolved, or starred, a label added or removed, or an incoming webhook from an external system. From the trigger you branch True/False on conditions like channel, labels, or message text, pause with delay nodes, and run actions — then test with sample data and watch every run in the log.