forked from hesabix/arc
Babak Alizadeh
0c4079d765
Replace regex/LLM shims for continue/stop decisions with structural evidence (API tool_calls, Harmony parser, ObservationStore). Add DB regression fixtures and remove unused aux.agent_continuation prompt. Co-authored-by: Cursor <cursoragent@cursor.com>
42 lines
1.3 KiB
Python
42 lines
1.3 KiB
Python
"""Regression fixtures for Plan C evidence-based agent loop."""
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
from pathlib import Path
|
|
|
|
import pytest
|
|
|
|
from app.services.ai.ai_agent_continuation import assess_text_round_evidence
|
|
from app.services.ai.ai_exploration_service import ObservationStore, ThoughtRecord
|
|
|
|
_FIXTURES = Path(__file__).parent / "fixtures" / "agent_regression_v2.json"
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
def regression_cases():
|
|
return json.loads(_FIXTURES.read_text(encoding="utf-8"))
|
|
|
|
|
|
@pytest.mark.parametrize("case", json.loads(_FIXTURES.read_text(encoding="utf-8")), ids=lambda c: c["id"])
|
|
def test_agent_regression_v2(case):
|
|
store = ObservationStore()
|
|
if case.get("has_high_confidence_thought"):
|
|
store.add_thought(
|
|
ThoughtRecord(
|
|
thought_id="t1",
|
|
bundle_id="b1",
|
|
iteration=1,
|
|
body_markdown="### یافتهها",
|
|
confidence="high",
|
|
open_questions=[],
|
|
)
|
|
)
|
|
|
|
result = assess_text_round_evidence(
|
|
round_text=case["round_text"],
|
|
user_query=case["user_query"],
|
|
observation_store=store,
|
|
needs_tools=case["needs_tools"],
|
|
)
|
|
assert result.should_continue == case["expect_continue"]
|
|
assert result.goal_reached == case["expect_goal_reached"]
|