- description
- Start your agents: the roles, and three easy ways to run them (P-1116; options TBD)
- tags
- #onboarding #inherited #wip
Start your agents¶
Each agent is an AI working in one role on your project, in the background. Start one per role; each registers itself and shows up on Agents as available.
| Role | What it does |
|---|---|
| designer | Turns your thoughts into the spec, design, mockups and tasks |
| coder | Builds the tasks from the pipeline (optional on Pro) |
| dispatcher | Runs the coders, batch after batch (Master) |
| tester | Checks every release (Master) |
Three ways to run them¶
a. A prompt to paste. Any AI chat that can reach the web: paste its start prompt and leave the chat open. Simplest; it works only while the chat runs.
b. Claude Code on your Mac. Install Claude Code, open a terminal in an empty folder, and start it with the role's start prompt. It keeps working in the background.
c. An agentic setup with a config file. A few easy setups, each with a ready config file to download. Coming: we're picking the setups.
The start prompt and your agent's token are on AI tokens: make a token for the role and copy the start line it shows.