Skill

The skill provides functionality to the agent, while masking the intricacies of its underlying implementation. Multiple skills can be connected to an agent, from which the agent can choose to achieve its objective.

Blocks and edges within a skill compose a sub chain within the chain of the application. We will refer to this as the skill chain. The skill chain is only run when an agent chooses to employ a skill. Keep in mind that a skill chain can be run multiple times within a single execution of an agent.

The Agent Input block within the skill block represents the inputs that an agent will populate when employing a skill. The structure of these inputs are defined by the skill block.

The value that is passed into the Skill Output block will be the outcome of employing the skill. At this point, the skill will relinquish execution control back to the agent.

When constructing skill chains, a few rules apply in addition to the rules regarding application chains. You can read about them in the Rules section.

Skill Settings

The name, description, and agent inputs are provided to the connected agent. The agent will use the name and description of its connected skills when deciding which skill to employ. Once it has chosen a skill, it will use the agent inputs information to fill in values for each input, which will be available to use within the skill chain.

Return Directly

If you would like the agent to immediately output whatever the skill outputs, you can check the return directly checkbox.

You can make an agent function essentially as a router by checking return directly for each of its skills.