Your First Agent and Skill

You program a robot with two things:

• An agent is a prompt plus a list of skills. An AI model watches the camera, listens to you, and picks a skill when it fits.
• A skill is something the robot can do. It’s a Python class.

On this page you create an agent in the web app, look at the Python it generates, write your own skill, and watch the agent use it. It takes about five minutes.

​

1. Open the robot and your editor

​

2. Create an agent

1. On the Agent page, open the agent picker at the top left and choose Create agent.
2. On the Identity tab, name it Cheerful Agent and enter the prompt: You are an enthusiastic robot. Celebrate good news physically.
3. On the Skills tab, click Add skill, pick Wave and HeadEmotion, and click Create.
4. Press Start and send “Hi! What can you do?” in the chat. The agent answers and picks a skill on its own.

Chatting needs an AI backend: your own Gemini key or the Innate service key. In the simulator you choose it during setup.

​

3. Look at the file it made

The form saved your agent as a normal Python file, workspace/custom_agents/cheerful_agent.py. Open it in your editor: That’s all an agent is: a name, a prompt in get_prompt(), and a list of skills in get_skills(). The form also turned on the microphone (get_inputs()) and made the robot look at the person it talks to (uses_gaze()). Every option is on the Agents page.

​

4. Write your own skill

1. Create workspace/custom_skills/victory_spin.py:
   from innate import Mobility, Skill, SkillReturn
   
   
   class VictorySpin(Skill):
       mobility: Mobility  # injected; guaranteed before execute() starts
   
       def guidelines(self) -> str:
           return "Use to celebrate: the robot spins in place once."
   
       def execute(self, spins: int = 1) -> SkillReturn:
           """Spin the robot in place to celebrate.
   
           Args:
               spins: How many full turns to make (1-3).
           """
           for _ in range(2 * min(int(spins), 3)):
               self.mobility.rotate(3.14)  # half turn, blocking; two = one spin
           return "Spun with joy"
   
2. Save. That’s the whole deploy: the robot loads the skill a second or two later, with no build or restart.
3. Open the skill menu in the web app and run victory_spin. The robot spins.

The agent reads guidelines() to decide when to use the skill, and the execute() signature and docstring to know how to call it. Edit the file, save, and run it again: that’s the whole loop.

​

5. Give it to your agent

1. Select Cheerful Agent, open the Skills tab, add VictorySpin, and click Save.
2. Tell the agent “I just merged a big PR!”. It decides on its own that good news calls for a victory spin.

You’ve built a robot app: a prompt, three skills, and one of them your own code.

​

When changes take effect

​

Next steps