Learning Path

The smallest useful agent is a loop that calls the model, runs tools, and feeds results back.

The loop stays stable while capabilities register into a dispatch table.

Dangerous actions need a harness decision point before the shell runs.

Cross-cutting behavior belongs around the loop, not tangled inside it.

Routing tasks by category decouples intent from implementation.

Subagents give each subtask a clean context while preserving the main thread.

Inject specialized knowledge only when the task actually needs it.

A well-designed hook system makes the agent infinitely extensible.

The agent can keep reasoning while slow work completes elsewhere.

A single config file controls the entire agent harness behavior.