File size: 2,611 Bytes
d754d1e
 
 
a140732
 
d754d1e
 
df7309c
d754d1e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a140732
d754d1e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cfa210f
df7309c
 
a140732
 
 
 
 
 
 
 
 
 
 
 
cfa210f
a140732
 
 
 
df7309c
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
import importlib
import sys

from support_triage_env.environment import SupportTriageCoreEnvironment


def _load_inference(monkeypatch):
    monkeypatch.delenv("OPENAI_API_KEY", raising=False)
    monkeypatch.setenv("HF_TOKEN", "test-token")
    monkeypatch.setenv("API_BASE_URL", "https://api.openai.com/v1")
    monkeypatch.setenv("MODEL_NAME", "test-model")
    sys.modules.pop("inference", None)
    return importlib.import_module("inference")


def test_parse_action_accepts_wrapped_json(monkeypatch):
    inference = _load_inference(monkeypatch)

    action = inference._parse_action(
        "Use this action:\n```json\n{\"action_type\":\"submit\",\"value\":null}\n```"
    )

    assert action.action_type == "submit"
    assert action.value is None


def test_rule_based_fallback_solves_all_default_tasks(monkeypatch, capsys):
    inference = _load_inference(monkeypatch)

    def _raise_provider_error(prompt: str):
        del prompt
        raise RuntimeError("provider unavailable")

    monkeypatch.setattr(inference, "_request_action_with_tools", _raise_provider_error)
    monkeypatch.setattr(inference, "_request_action_with_json", _raise_provider_error)

    for task_id in ("billing_duplicate_charge", "shipping_damaged_replacement", "account_takeover_fraud"):
        success, rewards = inference._run_task(task_id)
        output = capsys.readouterr().out

        assert success is True
        assert rewards
        assert f"[START] task={task_id} env=support_triage_env model=test-model" in output
        assert "[END] success=true" in output
        assert "score=0.99" in output


def test_strategy_handles_seeded_billing_variant(monkeypatch):
    inference = _load_inference(monkeypatch)
    env = SupportTriageCoreEnvironment()
    observation = env.reset(task_id="billing_duplicate_charge", seed=1)

    while not observation.done:
        action = inference._planned_action(observation)
        result = env.step(action)
        observation = result.observation

    state = env.state()
    assert state.submitted is True
    assert 0.98 < state.completion_score < 1.0
    assert state.workspace.status == "resolved"
    assert "auth_hold" in state.workspace.tags


def test_inference_accepts_openai_api_key(monkeypatch):
    monkeypatch.delenv("HF_TOKEN", raising=False)
    monkeypatch.setenv("OPENAI_API_KEY", "sk-test")
    monkeypatch.setenv("API_BASE_URL", "https://api.openai.com/v1")
    monkeypatch.setenv("MODEL_NAME", "test-model")
    sys.modules.pop("inference", None)

    inference = importlib.import_module("inference")

    assert inference.API_KEY == "sk-test"