Expand description
Executor trait boundary and pure eval orchestration.
The EvalExecutor trait isolates the harness orchestration from the
concrete agent runner. This is the “interface guard rail”: run_suite
depends only on the trait, so it can be unit-tested with a fake executor
and the production executor implementation can be swapped
without touching the orchestration logic.
Traits§
- Eval
Executor - Executes a single eval task attempt and returns the verified outcome.
Functions§
- run_
suite - Pure orchestration core: loop tasks × attempts through the executor, compute per-task metrics, and assemble the report.