Module conformance 

A conformance harness for AgentExecutor implementations.

The TCK grades servers. Nothing graded the thing an adopter actually writes. The paths they get wrong are the awkward ones — cancellation arriving mid-work, a parked task reported as an error, an event emitted after a terminal state — and those are exactly the cases people skip when writing tests by hand.

This drives an executor directly against a real event queue: no server, no ports, no model. It grades protocol invariants, the ones that hold for any agent whatever it does, and deliberately says nothing about whether the agent is any good at its job.

#[tokio::test]
async fn my_executor_is_conformant() {
    let report = a2a_protocol_server::conformance::check(Arc::new(MyExecutor)).await;
    report.assert_pass();
}

How it grades

Three outcomes, and the middle one is the one that keeps the score honest. A check that did not apply — because the executor never did the thing it is about — is not graded, and a report that grades nothing fails rather than passing vacuously. Both rules are the ones tck/ already follows, for the reason its own README gives: a run that measured nothing once reported full marks.

What it cannot tell you

It runs each check once, so it cannot find a race. It supplies its own message, so an executor that only misbehaves on particular input will pass — give it that input with check_with. It supplies its own CallContext, so an executor that reads the caller’s tenant or identity needs check_with_context to be graded against the one it will really meet. It bounds each drive at 30 seconds and grades an executor that overruns rather than waiting on it, so a very slow but correct executor is reported as a failure. And it grades the executor, not the server: the TCK is still what says your deployment conforms.

Structs

CheckResult

One graded check.

Report

Everything the harness found.

Enums

Outcome

How one check came out.

Functions

check

Grades an executor against a default one-line text message.

check_with

Grades an executor against a message you supply.

check_with_context

Grades an executor with a CallContext you supply.