87 lines
3.7 KiB
Python
87 lines
3.7 KiB
Python
import os
|
||
import unittest
|
||
|
||
from app.services.ai_agent_kernel import (
|
||
compute_capability_score,
|
||
gate_execution,
|
||
resolve_correction,
|
||
select_intent_route,
|
||
)
|
||
|
||
|
||
class AIAgentKernelTest(unittest.TestCase):
|
||
def test_select_intent_route_prefers_complex_for_multi_clause_instruction(self) -> None:
|
||
decision = select_intent_route(
|
||
"给我新增个项目，openclaw自动化部署，为了后续方便统一发布。",
|
||
project_names=["OpenClaw 自动化部署"],
|
||
)
|
||
self.assertEqual(decision.function_name, "classify_intent_complex")
|
||
self.assertEqual(decision.tier, "high")
|
||
self.assertTrue(decision.is_complex)
|
||
|
||
def test_select_intent_route_simple_instruction(self) -> None:
|
||
decision = select_intent_route("归档项目#51")
|
||
self.assertEqual(decision.function_name, "classify_intent_ai_pmo_kernel")
|
||
self.assertEqual(decision.tier, "low")
|
||
self.assertFalse(decision.is_complex)
|
||
|
||
def test_resolve_correction_detects_rollback(self) -> None:
|
||
result = resolve_correction("这条不是新增项目，回滚刚才那次操作。")
|
||
self.assertIn(result.action, {"rollback", "replace_target"})
|
||
self.assertGreaterEqual(result.confidence, 0.78)
|
||
self.assertTrue(result.matched_keywords)
|
||
|
||
def test_gate_execution_respects_threshold_and_risk(self) -> None:
|
||
old_auto = os.environ.get("AGENT_GATE_AUTO_THRESHOLD")
|
||
old_review = os.environ.get("AGENT_GATE_REVIEW_THRESHOLD")
|
||
old_review_below_auto = os.environ.get("AGENT_GATE_REVIEW_BELOW_AUTO")
|
||
os.environ["AGENT_GATE_AUTO_THRESHOLD"] = "0.82"
|
||
os.environ["AGENT_GATE_REVIEW_THRESHOLD"] = "0.60"
|
||
try:
|
||
low = gate_execution(confidence=0.42, risk_level="low")
|
||
self.assertEqual(low.action, "reject")
|
||
pending = gate_execution(confidence=0.70, risk_level="medium")
|
||
self.assertEqual(pending.action, "auto_execute")
|
||
high_risk = gate_execution(confidence=0.86, risk_level="high")
|
||
self.assertEqual(high_risk.action, "pending_manual")
|
||
allowed = gate_execution(confidence=0.93, risk_level="high")
|
||
self.assertEqual(allowed.action, "auto_execute")
|
||
os.environ["AGENT_GATE_REVIEW_BELOW_AUTO"] = "1"
|
||
pending_with_switch = gate_execution(confidence=0.70, risk_level="medium")
|
||
self.assertEqual(pending_with_switch.action, "pending_manual")
|
||
forced = gate_execution(
|
||
confidence=0.99,
|
||
risk_level="low",
|
||
force_manual=True,
|
||
force_manual_code="intent_conflict_write",
|
||
force_manual_reason="规则与LLM冲突",
|
||
)
|
||
self.assertEqual(forced.action, "pending_manual")
|
||
self.assertEqual(forced.gate_code, "intent_conflict_write")
|
||
finally:
|
||
if old_auto is None:
|
||
os.environ.pop("AGENT_GATE_AUTO_THRESHOLD", None)
|
||
else:
|
||
os.environ["AGENT_GATE_AUTO_THRESHOLD"] = old_auto
|
||
if old_review is None:
|
||
os.environ.pop("AGENT_GATE_REVIEW_THRESHOLD", None)
|
||
else:
|
||
os.environ["AGENT_GATE_REVIEW_THRESHOLD"] = old_review
|
||
if old_review_below_auto is None:
|
||
os.environ.pop("AGENT_GATE_REVIEW_BELOW_AUTO", None)
|
||
else:
|
||
os.environ["AGENT_GATE_REVIEW_BELOW_AUTO"] = old_review_below_auto
|
||
|
||
def test_compute_capability_score(self) -> None:
|
||
score = compute_capability_score(
|
||
intent_accuracy=0.82,
|
||
gate_precision=0.80,
|
||
exec_success=0.78,
|
||
correction_success=0.76,
|
||
)
|
||
self.assertAlmostEqual(score, 0.795, places=3)
|
||
|
||
|
||
if __name__ == "__main__":
|
||
unittest.main()
|