name	evaluate-lunchtable-agent
description	Use when testing, comparing, or hardening Lunch Table Games gameplay agents.

Evaluate Lunch Table Agent

Use this skill when validating that a gameplay agent can participate fairly, deterministically, and without hidden authority.

Run the focused legal-action and parity tests for the changed package.
Run self-play tests or deterministic replay tests for the changed game family.
Check that every selected actionId exists in the current legal action catalog.
Check that private state appears only in scoped owner views.
Compare replay, self-play, or benchmark outputs after rules or policy changes.

A stronger model may reason better, but it does not get more game authority.
Invalid external action ids must fail loudly.
Evaluation fixtures should use deterministic seeds and stable timestamps.
Advisory agents can explain or coach, but they do not submit live gameplay intents.

name	evaluate-lunchtable-agent
description	Use when testing, comparing, or hardening Lunch Table Games gameplay agents.

Use this skill when validating that a gameplay agent can participate fairly, deterministically, and without hidden authority.

Run the focused legal-action and parity tests for the changed package.
Run self-play tests or deterministic replay tests for the changed game family.
Check that every selected actionId exists in the current legal action catalog.
Check that private state appears only in scoped owner views.
Compare replay, self-play, or benchmark outputs after rules or policy changes.

A stronger model may reason better, but it does not get more game authority.
Invalid external action ids must fail loudly.
Evaluation fixtures should use deterministic seeds and stable timestamps.
Advisory agents can explain or coach, but they do not submit live gameplay intents.

evaluate-lunchtable-agent