# retoor <retoor@molodetz.nl>

from devplacepy.services.devii.agents.actions import AGENT_ACTIONS
from devplacepy.services.devii.agents.evals import EVALS
from devplacepy.services.devii.agents.policy import CATEGORIES
from devplacepy.services.devii.registry import CATALOG


def test_twenty_unique_evals():
    assert len(EVALS) == 20
    assert len({item["uid"] for item in EVALS}) == 20


def test_evals_cover_channels_and_kinds():
    channels = {item["channel"] for item in EVALS}
    assert channels == {"main", "telegram"}
    kinds = {item["expect"] for item in EVALS}
    assert "direct" in kinds
    assert kinds - {"direct"} <= set(CATEGORIES)


def test_evals_are_well_formed_and_grounded():
    known = set(CATALOG.by_name()) | {a.name for a in AGENT_ACTIONS}
    for item in EVALS:
        assert item["prompt"].strip()
        assert item["end_state"].strip()
        assert item["channel"] in ("main", "telegram")
        assert item["expect"] == "direct" or item["expect"] in CATEGORIES
        for tool in item["tools"]:
            assert tool in known
        if item["expect"] != "direct":
            assert "agent_start" in item["tools"]
            assert {"agent_result", "agent_status", "agent_cancel"} & set(
                item["tools"]
            )