feat: Phase 2 complete — 13 Phases of COBOL type classification and test benchmark

P0.6: gcov infrastructure
P1: extract_structure output expansion (11 new feature fields)
P2: Confusion group rule engine (8 pairs + contradiction + backtrack)
P3: 4-factor confidence calculation + quality gate update
P4: 33+2 COBOL program type test samples (22 files, 7 categories)
P5: parametrized/ test data generation engine
P6: japanese_data.py lookup tables
P7-10: Type-specific test suites (~159 parametrized tests)
P11: Full classification pipeline (classify_program) + orchestrator integration
P12: Documentation (module-interfaces, test-plan v3.0, coverage-matrix)

Architecture decisions:
- classification_pipeline/ merged to hina/pipeline/
- parametrized/ as independent module
- japanese_data.py as root-level file
- hina/__all__ only exports classify_program()

Co-Authored-By: Claude <noreply@anthropic.com>

agents/__init__.py

@@ -0,0 +1,22 @@
+"""LLM 智能体包
+
+公开 API:
+  LLMClient          — LLM API 客户端（含缓存 + 重试）
+  Agent1Parser       — COPYBOOK → FieldTree
+  Agent2Data         — FieldTree → TestSuite（测试数据设计）
+  Agent3Diagnostic   — FieldResult → 诊断建议文本
+"""
+
+from __future__ import annotations
+
+from .llm import LLMClient
+from .agent1_parser import Agent1Parser
+from .agent2_data import Agent2Data
+from .agent3_diagnostic import Agent3Diagnostic
+
+__all__ = [
+    "LLMClient",          # class
+    "Agent1Parser",       # class
+    "Agent2Data",         # class
+    "Agent3Diagnostic",   # class
+]

agents/llm.py

@@ -15,7 +15,12 @@ class LLMClient:
 
     def _get(self, k):
         p = self.dir / f"{k}.json"
-        return json.loads(p.read_text())["response"] if p.exists() else None
+        if not p.exists():
+            return None
+        try:
+            return json.loads(p.read_text())["response"]
+        except (json.JSONDecodeError, KeyError):
+            return None
 
     def _set(self, k, v):
         (self.dir / f"{k}.json").write_text(json.dumps({"response": v}))