agents · semantic context
Give the agent a program model, not just source text
2ot already has semantic-agent primitives for repair context, callable signatures, workspace graphs and blast radius. They let a harness bound a change by program relationships instead of relying on grep and prompt context alone.
Existing semantic primitives
repair context
callable signature context
semantic graph
workspace graph
blast radius
compact LLM diagnostics
These are compiler/agent-context capabilities. A full physical plant graph spanning IO, drives, HMI and observed machine state is a broader product direction, not a current claim.
Definition relationships
FB_Conveyor
├─ reads → SafetyOK, Photoeye_01
├─ writes → MotorStart, Fault
└─ calls → FB_Drive
readswritescallstypesaddressesdepends_on
BLAST RADIUS
Ask what a change can affect
Before a repair is applied, semantic dependency information can help identify the definitions around the changed symbol. That is useful both for agent context selection and for the planned transaction layer.
Example question
“What logic could be affected if
SafetyOK changes?”SafetyOK → FB_Conveyor → MotorStart → callers / dependents