Most people working with a coding agent are guessing. You type, something comes back, and when it goes wrong the only move you have is rephrasing and hoping.

This is a 50-page manual about the part nobody explains: what actually gets assembled and sent to the model on every turn, and which parts of that you control.

1. The loop, and what the model sees. How a turn is built, why the same request behaves differently twice, and why context is a budget rather than a container.
2. The instruction file. The cheapest lever you have, and the one most often bloated. Generate it, then prune it, and keep pruning.
3. Planning and verification. Read the plan properly, put something in the loop that can tell the agent it is wrong, and let it interview you first.
4. Packaged workflows. When a prompt should stop being a prompt, why only the description costs you anything, and how to tell if it helps.
5. Lifecycle hooks. The only deterministic control you have, and the one hook worth building first.
6. Sub-agents and teams. What delegation really costs, when it wins, and the honest answer about when a team earns its coordination.
7. Distribution and habits. Choosing a model per task, sharing a setup, and what survives a busy week.

Plus six worksheets for your own project, a symptom-to-cause troubleshooting table, a glossary, and a two-week plan.

No framework, and no tool-specific screenshots to go stale. Just the mental model and the habits.