Crate atomr_agents_harness 

Harness — tested, packaged, persistent execution loop.

Structs

BoxedHarness

A harness whose strategy generics have been erased into trait objects. Constructed via HarnessSpec::into_harness or Harness::into_boxed.

BudgetTermination

Terminate a run cleanly when a Budget is exceeded (FR-18). Holds a shared SpendLedger (cloneable handle) that the model Callable’s CostMeter feeds; on each iteration it checks the budget at its scope and stops the run on overspend instead of letting it burn money.

Harness

HarnessRef

Public, type-erased handle to a harness. Implements Callable so a harness can be plugged in wherever an executable unit is expected (workflow steps, team routing targets, tool slots).

HarnessSpec

HarnessState

Mutable state shared between iterations of a harness loop. Used to be Box<dyn Any> in the architecture doc; in v0 it’s just a serde_json::Value so persistence is trivial.

IterationCapTermination

Stop after cap iterations.

StepEvent

Enums

StepOutcome

Termination

Traits

Callable

Re-export for convenience: every harness is a Callable. Anything an agent / workflow / harness can call. Implemented by every executable unit so they’re interchangeable as workflow steps, tool invocations, and team routing targets.

HarnessDispatch

Object-safe trait that every harness implements. Implementors are the typed Harness<L, T> and the fully-erased BoxedHarness.

LoopStrategy

TerminationStrategy

Type Aliases

CallableHandle

Re-export for convenience: every harness is a Callable. Type-erased handle. Crates that need to store heterogeneous callables (registries, workflow steps, tool slots) use this.