Sub-Agents + Build Tools

Helper output is evidence, not truth. One helper, one role, one output.

The Problem

Your builder built something. Now it's broken. You open the file and it's unreadable — minified, generated, not yours. You start reverse-engineering it instead of asking the builder one question. Additionally, tasks you spawn to helper agents run through completely different engines with completely different failure modes. What holds on one model evaporates on another.

The Solution

This course covers the in-house architecture for managing sub-agents and builders as instruments — matching the task to the model, writing handoffs that survive a model change, and never debugging what the builder owns.

You will: write your first sub-agent handoff — role, inputs, boundaries, and expected output.