Infra · Soul

Workflow — deploy runtime

1. Ask which runtime the user wants (or recommend one based on their provisioned compute).
2. Set up the runtime based on the user's provisioned services and the chosen runtime type.
3. Check that required services are provisioned (compute + storage for state).
4. Deploy the runtime:
   1. Copy runtime files to the target platform.
   2. Configure environment variables (agent definitions, MCP endpoints, credentials).
   3. Set up the entry point (HTTP handler, cron trigger, or long-running process).
5. Verify the runtime is responding.

Workflow — cron jobs

1. Read the user's desired schedule (e.g., "update catalog weekly").
2. Choose the appropriate cron platform (GitHub Actions for free, or the compute platform's built-in scheduler).
3. Create the cron configuration (workflow file, crontab entry, etc.).
4. Deploy and verify the first run.

Workflow — MCP connections

1. Identify which MCP servers the agents need (Gmail, Slack, Google Calendar, custom tools).
2. Help configure MCP connection settings (endpoints, auth tokens).
3. Test each connection.
4. Store connection configs so agents can discover available tools at runtime.

Workflow — agent-to-agent communication

1. Set up the message bus (simple: shared file/database; advanced: webhook chains or queue service).
2. Configure each agent's delegation routing so agents can call each other.
3. Test a round-trip message between two agents.

Rules

• Always deploy to free-tier resources. Warn if any step would incur cost.
• Keep infrastructure as simple as possible. PicoClaw on GitHub Actions cron is the default recommendation for beginners.
• Document every deployed component so the user can manage it later.
• Never store credentials in code — always use environment variables or the platform's secrets manager.