Context engineering laboratory

Compile the model’s working state

Prompt engineering shapes an instruction. Context engineering—sometimes casually called contextual engineering—decides the complete token state the model receives at each inference step: instructions, evidence, tool schemas, skills, memory, conversation history, and recent tool results.

Use the context compiler to add, remove, and prioritize real categories of state. The model’s weights never change. What changes is the finite working set from which the next action is sampled.

A finite illuminated context chamber being packed with instructions, evidence, tools, history, skills, and memory
Context engineering · Assemble the finite working state
Manipulate
Budget · inclusion · priority · provenance
Observe
Packed tokens · overflow · unused headroom
Never shown
A fake context-quality percentage
Context compiler3,000 / 3,000 fixed teaching tokens packed

Available state

Model context at turn t100% full
System policy310 t
User goal190 t
Tool schemas640 t
Activated skill520 t
Six event sources980 t
Durable memory360 t
Highest-authority instructions firstData keeps its provenance
Selected6 blocks
Excluded1 blocks
Model weightsunchanged
Recompiledevery turn

Primary research

Inspect the mechanism at its source.

Project architecture and terminology were checked against official documentation on August 10, 2026. Research papers are linked to their original publication records.

Finite attention and just-in-time retrievalEffective context engineeringOriginal source ↗What enters a real agent contextOpenClaw contextOriginal source ↗Stable, context, and volatile tiersHermes prompt assemblyOriginal source ↗Progressive disclosureAgent Skills specificationOriginal source ↗

Keep experimenting

Change the layer, keep the system visible.

AI Lab indexSee every interactive laboratoryBrowse the suite →Transformer microscopeGo beneath the agent runtime into the modelInspect the matrices →