Implement MetricsStore for project-scoped agent metrics.

Add ADR-004 storage layer with append-only executions, summary
regeneration, idempotency keys, and retention pruning. Wire memory init
to scaffold .kaizen/metrics/ by default and add unit tests.

tests/test_metrics.py

@@ -0,0 +1,107 @@
+"""Tests for project-scoped metrics storage (ADR-004)."""
+
+from __future__ import annotations
+
+import json
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+
+import pytest
+
+from kaizen_agentic.metrics import MetricsStore, DEFAULT_RETENTION_DAYS
+
+
+def _old_timestamp(days: int) -> str:
+    dt = datetime.now(timezone.utc) - timedelta(days=days)
+    return dt.strftime("%Y-%m-%dT%H:%M:%SZ")
+
+
+@pytest.fixture
+def project_dir(tmp_path: Path) -> Path:
+    root = tmp_path / "demo-project"
+    root.mkdir()
+    return root
+
+
+class TestMetricsStore:
+    def test_scaffold_creates_directory_and_empty_executions(self, project_dir: Path):
+        store = MetricsStore(project_dir, "tdd-workflow")
+        path = store.scaffold()
+
+        assert path == project_dir / ".kaizen" / "metrics" / "tdd-workflow"
+        assert store.executions_path.exists()
+        assert store.executions_path.read_text() == ""
+
+    def test_append_and_read_executions(self, project_dir: Path):
+        store = MetricsStore(project_dir, "tdd-workflow")
+
+        assert store.append({"success": True, "quality_score": 0.9}) is True
+        assert store.append({"success": False, "execution_time_s": 12.5}) is True
+
+        records = store.read_executions()
+        assert len(records) == 2
+        assert records[0]["agent"] == "tdd-workflow"
+        assert records[0]["success"] is True
+        assert "timestamp" in records[0]
+
+    def test_idempotency_key_rejects_duplicate(self, project_dir: Path):
+        store = MetricsStore(project_dir, "coach")
+
+        assert store.append({"success": True}, idempotency_key="sess-1") is True
+        assert store.append({"success": True}, idempotency_key="sess-1") is False
+        assert len(store.read_executions()) == 1
+
+    def test_write_summary_regenerates_summary_json(self, project_dir: Path):
+        store = MetricsStore(project_dir, "tdd-workflow")
+        store.append({"success": True, "quality_score": 0.8, "execution_time_s": 10})
+        store.append({"success": True, "quality_score": 1.0, "execution_time_s": 20})
+
+        summary = store.write_summary()
+
+        assert summary["execution_count"] == 2
+        assert summary["success_rate"] == 1.0
+        assert summary["avg_quality_score"] == 0.9
+        assert summary["avg_execution_time_s"] == 15.0
+        assert store.summary_path.exists()
+        on_disk = json.loads(store.summary_path.read_text())
+        assert on_disk["execution_count"] == 2
+
+    def test_prune_removes_expired_records(self, project_dir: Path):
+        store = MetricsStore(project_dir, "tdd-workflow", retention_days=30)
+        store.scaffold()
+
+        old = {
+            "timestamp": _old_timestamp(45),
+            "agent": "tdd-workflow",
+            "success": False,
+        }
+        recent = {
+            "timestamp": _old_timestamp(1),
+            "agent": "tdd-workflow",
+            "success": True,
+            "quality_score": 0.7,
+        }
+        with store.executions_path.open("w", encoding="utf-8") as handle:
+            handle.write(json.dumps(old) + "\n")
+            handle.write(json.dumps(recent) + "\n")
+
+        removed = store.prune()
+
+        assert removed == 1
+        records = store.read_executions()
+        assert len(records) == 1
+        assert records[0]["success"] is True
+        summary = store.read_summary()
+        assert summary is not None
+        assert summary["execution_count"] == 1
+
+    def test_list_agents_with_metrics(self, project_dir: Path):
+        MetricsStore(project_dir, "tdd-workflow").scaffold()
+        MetricsStore(project_dir, "coach").append({"success": True})
+
+        agents = MetricsStore.list_agents(project_dir)
+
+        assert agents == ["coach", "tdd-workflow"]
+
+    def test_default_retention_matches_adr(self):
+        assert DEFAULT_RETENTION_DAYS == 180