Context Cache

Keep the next action’s evidence inside a limited working set.

Pick the notes the next tool call needs, stay within five context slots, then run the step.

How to play

Each mission has five notes and room for only five context slots. Pick the facts the next tool call needs, then run it. A correct pair advances the trajectory; irrelevant notes consume space. The slot costs are simplified game units.

The systems concept

An agent’s context is an active working set. When a long trajectory must be compacted, preserving the evidence and authority needed by the next action matters more than retaining every earlier sentence. The game makes that selection visible as a budget decision.

Explore Volume III: Agentic Machine Learning Systems or return to the playground.