Crate zeroclaw_eval 

Agent evaluation harness for ZeroClaw.

Re-exports

pub use case::LlmTrace;

pub use case::TraceExpects;

pub use record::RunRecord;

pub use report::CaseReport;

pub use report::SuiteReport;

pub use runner::run_case;

pub use runner::run_suite;

Modules

case

The evaluation case format — JSON trace fixtures for deterministic replay.

grader

Grading: non-panicking checks over a RunRecord.

observer

An Observer that records tool-call outcomes and token usage from a run.

record

The artifact produced by running one eval case — what graders score.

replay

A ModelProvider that replays scripted LLM responses from an LlmTrace. Promoted from the test-only trace-replay helper so the same deterministic engine backs both the shipped zeroclaw eval command and the test suite.

report

Pass/fail aggregation and rendering.

runner

The runner: builds an isolated agent per case, drives it, and grades it.

tools

Deterministic built-in tools available to the replay agent.

Enums

Mode

How an evaluation suite is executed.