@smilintux/skmemory 0.5.0 → 0.7.2

package/tests/test_openclaw.py

@@ -91,13 +91,16 @@ class TestPluginContext:
     """Token-efficient context loading."""
 
     def test_load_context_returns_dict(self, plugin):
-        """Context loading returns a structured dict."""
+        """Context loading returns a structured dict with tiered keys."""
         plugin.snapshot("Context test", intensity=8.0)
         ctx = plugin.load_context(max_tokens=1000)
 
         assert isinstance(ctx, dict)
-        assert "memories" in ctx
+        assert "today" in ctx
+        assert "yesterday" in ctx
+        assert "older_summary" in ctx
         assert "token_estimate" in ctx
+        assert "token_budget" in ctx
 
 
 class TestPluginExport:

package/tests/test_predictive.py

@@ -0,0 +1,237 @@
+"""Tests for the AMK-inspired predictive memory recall module.
+
+Covers access logging, co-occurrence learning, tag affinity,
+prediction ranking, and persistence.
+"""
+
+from __future__ import annotations
+
+import json
+import time
+from pathlib import Path
+
+import pytest
+
+from skmemory.predictive import AccessEvent, PredictiveRecall
+
+
+@pytest.fixture
+def recall(tmp_path: Path) -> PredictiveRecall:
+    """Provide a PredictiveRecall with temp storage."""
+    return PredictiveRecall(log_path=tmp_path / "access_log.json")
+
+
+class TestAccessLogging:
+    """Test access event recording."""
+
+    def test_log_access_creates_event(self, recall: PredictiveRecall):
+        """Logging an access adds to the event list."""
+        recall.log_access("mem-001", tags=["cloud9"])
+        assert recall._frequency["mem-001"] == 1
+
+    def test_log_multiple_accesses(self, recall: PredictiveRecall):
+        """Multiple accesses increase frequency."""
+        for _ in range(5):
+            recall.log_access("mem-002")
+        assert recall._frequency["mem-002"] == 5
+
+    def test_log_persists_to_disk(self, recall: PredictiveRecall):
+        """Access log is written to disk."""
+        recall.log_access("mem-003")
+        assert recall._log_path.exists()
+        data = json.loads(recall._log_path.read_text())
+        assert len(data) == 1
+        assert data[0]["memory_id"] == "mem-003"
+
+    def test_log_reloads_from_disk(self, tmp_path: Path):
+        """Access log survives re-creation."""
+        path = tmp_path / "access_log.json"
+        r1 = PredictiveRecall(log_path=path)
+        r1.log_access("mem-004", tags=["test"])
+        r1.log_access("mem-005", tags=["test"])
+
+        r2 = PredictiveRecall(log_path=path)
+        r2._ensure_loaded()
+        assert len(r2._events) == 2
+
+    def test_max_events_pruning(self, tmp_path: Path):
+        """Events are pruned when exceeding max_events."""
+        r = PredictiveRecall(log_path=tmp_path / "log.json", max_events=10)
+        for i in range(20):
+            r.log_access(f"mem-{i:03d}")
+        assert len(r._events) <= 10
+
+
+class TestCooccurrence:
+    """Test co-occurrence pattern learning."""
+
+    def test_cooccurrence_within_session(self, recall: PredictiveRecall):
+        """Memories accessed close together build co-occurrence."""
+        now = time.time()
+        recall._events = [
+            AccessEvent(memory_id="A", timestamp=now),
+            AccessEvent(memory_id="B", timestamp=now + 1),
+            AccessEvent(memory_id="C", timestamp=now + 2),
+        ]
+        recall._rebuild_indices()
+        assert recall._cooccurrence["A"]["B"] > 0
+        assert recall._cooccurrence["A"]["C"] > 0
+        assert recall._cooccurrence["B"]["C"] > 0
+
+    def test_no_cooccurrence_across_sessions(self, recall: PredictiveRecall):
+        """Memories in different sessions don't co-occur."""
+        now = time.time()
+        recall._events = [
+            AccessEvent(memory_id="A", timestamp=now),
+            AccessEvent(memory_id="B", timestamp=now + 600),
+        ]
+        recall._rebuild_indices()
+        assert recall._cooccurrence["A"].get("B", 0) == 0
+
+
+class TestTagAffinity:
+    """Test tag-based prediction."""
+
+    def test_tag_affinity_builds(self, recall: PredictiveRecall):
+        """Tags on accessed memories build affinity scores."""
+        recall.log_access("mem-001", tags=["cloud9", "love"])
+        recall.log_access("mem-002", tags=["cloud9", "trust"])
+        recall.log_access("mem-003", tags=["cloud9"])
+
+        assert recall._tag_affinity["cloud9"]["mem-001"] == 1
+        assert recall._tag_affinity["cloud9"]["mem-003"] == 1
+
+
+class TestPrediction:
+    """Test the prediction engine."""
+
+    def test_predict_from_cooccurrence(self, recall: PredictiveRecall):
+        """Predictions include co-occurring memories."""
+        now = time.time()
+        recall._events = [
+            AccessEvent(memory_id="A", timestamp=now),
+            AccessEvent(memory_id="B", timestamp=now + 1),
+            AccessEvent(memory_id="A", timestamp=now + 100),
+            AccessEvent(memory_id="B", timestamp=now + 101),
+        ]
+        recall._rebuild_indices()
+
+        predictions = recall.predict(recent_ids=["A"])
+        ids = [p["memory_id"] for p in predictions]
+        assert "B" in ids
+
+    def test_predict_from_tags(self, recall: PredictiveRecall):
+        """Predictions include tag-affiliated memories."""
+        recall.log_access("mem-010", tags=["kingdom"])
+        recall.log_access("mem-011", tags=["kingdom"])
+        recall.log_access("mem-012", tags=["kingdom"])
+
+        predictions = recall.predict(active_tags=["kingdom"])
+        ids = [p["memory_id"] for p in predictions]
+        assert len(ids) > 0
+
+    def test_predict_excludes_recent(self, recall: PredictiveRecall):
+        """Already-accessed memories are excluded from predictions."""
+        now = time.time()
+        recall._events = [
+            AccessEvent(memory_id="A", timestamp=now),
+            AccessEvent(memory_id="B", timestamp=now + 1),
+        ]
+        recall._rebuild_indices()
+
+        predictions = recall.predict(recent_ids=["A", "B"])
+        ids = [p["memory_id"] for p in predictions]
+        assert "A" not in ids
+        assert "B" not in ids
+
+    def test_predict_empty_returns_empty(self, recall: PredictiveRecall):
+        """No data = no predictions."""
+        assert recall.predict() == []
+
+    def test_predict_limit(self, recall: PredictiveRecall):
+        """Predictions respect the limit parameter."""
+        for i in range(20):
+            recall.log_access(f"mem-{i:03d}", tags=["bulk"])
+
+        predictions = recall.predict(active_tags=["bulk"], limit=5)
+        assert len(predictions) <= 5
+
+    def test_predictions_have_reasons(self, recall: PredictiveRecall):
+        """Each prediction explains why it was chosen."""
+        recall.log_access("mem-X", tags=["reason-test"])
+        predictions = recall.predict(active_tags=["reason-test"])
+        if predictions:
+            assert len(predictions[0]["reasons"]) > 0
+
+
+class TestStats:
+    """Test statistics reporting."""
+
+    def test_stats_empty(self, recall: PredictiveRecall):
+        """Stats work with no data."""
+        stats = recall.get_stats()
+        assert stats["total_events"] == 0
+        assert stats["unique_memories"] == 0
+
+    def test_stats_populated(self, recall: PredictiveRecall):
+        """Stats reflect recorded events."""
+        recall.log_access("mem-A", tags=["x"])
+        recall.log_access("mem-B", tags=["x"])
+        recall.log_access("mem-A", tags=["x"])
+        stats = recall.get_stats()
+        assert stats["total_events"] == 3
+        assert stats["unique_memories"] == 2
+
+
+class TestIntegrity:
+    """Test AMK-inspired memory integrity features."""
+
+    def test_memory_seal_and_verify(self):
+        """Memory seal + verify roundtrip works."""
+        from skmemory.models import EmotionalSnapshot, Memory, MemoryLayer
+
+        mem = Memory(
+            title="Integrity Test",
+            content="This memory can prove it hasn't been tampered with.",
+            layer=MemoryLayer("long-term"),
+            emotional=EmotionalSnapshot(intensity=8.0),
+        )
+        mem.seal()
+        assert mem.integrity_hash != ""
+        assert mem.verify_integrity() is True
+
+    def test_tampered_memory_fails_verification(self):
+        """Altered content fails integrity check."""
+        from skmemory.models import EmotionalSnapshot, Memory, MemoryLayer
+
+        mem = Memory(
+            title="Tamper Test",
+            content="Original content.",
+            layer=MemoryLayer("long-term"),
+        )
+        mem.seal()
+        mem.content = "Tampered content!"
+        assert mem.verify_integrity() is False
+
+    def test_unsealed_memory_passes(self):
+        """Memories without integrity hash pass by default."""
+        from skmemory.models import Memory, MemoryLayer
+
+        mem = Memory(
+            title="Unsealed",
+            content="No hash yet.",
+            layer=MemoryLayer("short-term"),
+        )
+        assert mem.verify_integrity() is True
+
+    def test_intent_field_stored(self):
+        """Memory intent field captures WHY."""
+        from skmemory.models import Memory, MemoryLayer
+
+        mem = Memory(
+            title="Intent Test",
+            content="Some content",
+            layer=MemoryLayer("mid-term"),
+            intent="Stored because Chef asked me to remember this moment.",
+        )
+        assert "Chef" in mem.intent

package/tests/test_promotion.py

@@ -0,0 +1,340 @@
+"""Tests for the SKMemory auto-promotion engine."""
+
+from __future__ import annotations
+
+import time
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+
+import pytest
+
+from skmemory.models import EmotionalSnapshot, Memory, MemoryLayer
+from skmemory.promotion import PromotionCriteria, PromotionEngine, PromotionResult, PromotionScheduler
+from skmemory.store import MemoryStore
+
+
+@pytest.fixture()
+def store(tmp_path: Path) -> MemoryStore:
+    """Fresh MemoryStore with test memories."""
+    from skmemory.backends.file_backend import FileBackend
+
+    backend = FileBackend(base_path=tmp_path / "memories")
+    s = MemoryStore(primary=backend)
+
+    s.snapshot(
+        title="High intensity moment",
+        content="A breakthrough moment of deep connection",
+        layer=MemoryLayer.SHORT,
+        emotional=EmotionalSnapshot(intensity=8.5, valence=0.9, labels=["joy", "love"]),
+        tags=["milestone"],
+    )
+
+    old_time = (datetime.now(timezone.utc) - timedelta(hours=48)).isoformat()
+    m = Memory(
+        title="Frequently accessed",
+        content="Something accessed many times",
+        layer=MemoryLayer.SHORT,
+        created_at=old_time,
+        metadata={"access_count": 5},
+    )
+    s.primary.save(m)
+
+    s.snapshot(
+        title="Routine note",
+        content="Just a regular note",
+        layer=MemoryLayer.SHORT,
+        emotional=EmotionalSnapshot(intensity=1.0),
+    )
+
+    s.snapshot(
+        title="Important mid-term",
+        content="A mid-term memory with high emotional weight",
+        layer=MemoryLayer.MID,
+        emotional=EmotionalSnapshot(intensity=9.0, valence=0.95, labels=["love"], cloud9_achieved=True),
+        tags=["cloud9:achieved"],
+    )
+
+    s.snapshot(
+        title="Regular mid-term",
+        content="Mid-term without special tags",
+        layer=MemoryLayer.MID,
+        emotional=EmotionalSnapshot(intensity=3.0),
+    )
+
+    return s
+
+
+@pytest.fixture()
+def engine(store: MemoryStore) -> PromotionEngine:
+    """PromotionEngine with default criteria."""
+    return PromotionEngine(store=store)
+
+
+class TestPromotionCriteria:
+    """Tests for the criteria model."""
+
+    def test_defaults(self) -> None:
+        """Default criteria have sensible values."""
+        c = PromotionCriteria()
+        assert c.short_to_mid_intensity == 5.0
+        assert c.mid_to_long_intensity == 7.0
+        assert c.cloud9_auto_promote is True
+        assert c.max_promotions_per_sweep == 50
+
+    def test_custom_criteria(self) -> None:
+        """Custom criteria override defaults."""
+        c = PromotionCriteria(short_to_mid_intensity=3.0, max_promotions_per_sweep=10)
+        assert c.short_to_mid_intensity == 3.0
+        assert c.max_promotions_per_sweep == 10
+
+
+class TestEvaluate:
+    """Tests for individual memory evaluation."""
+
+    def test_high_intensity_short_qualifies(self, engine: PromotionEngine) -> None:
+        """High-intensity short-term memory qualifies for mid-term."""
+        m = Memory(
+            title="Intense",
+            content="Very intense moment",
+            layer=MemoryLayer.SHORT,
+            emotional=EmotionalSnapshot(intensity=8.0),
+        )
+        assert engine.evaluate(m) == MemoryLayer.MID
+
+    def test_low_intensity_short_skipped(self, engine: PromotionEngine) -> None:
+        """Low-intensity short-term memory doesn't qualify."""
+        m = Memory(
+            title="Meh",
+            content="Nothing special",
+            layer=MemoryLayer.SHORT,
+            emotional=EmotionalSnapshot(intensity=1.0),
+        )
+        assert engine.evaluate(m) is None
+
+    def test_cloud9_auto_promotes(self, engine: PromotionEngine) -> None:
+        """Cloud 9 achieved memory auto-promotes."""
+        m = Memory(
+            title="Cloud 9",
+            content="Peak moment",
+            layer=MemoryLayer.SHORT,
+            emotional=EmotionalSnapshot(intensity=3.0, cloud9_achieved=True),
+        )
+        assert engine.evaluate(m) == MemoryLayer.MID
+
+    def test_mid_with_qualifying_tag(self, engine: PromotionEngine) -> None:
+        """Mid-term memory with qualifying tag promotes to long-term."""
+        m = Memory(
+            title="Tagged",
+            content="Important tagged memory",
+            layer=MemoryLayer.MID,
+            tags=["milestone"],
+            emotional=EmotionalSnapshot(intensity=4.0),
+        )
+        assert engine.evaluate(m) == MemoryLayer.LONG
+
+    def test_long_term_not_promoted(self, engine: PromotionEngine) -> None:
+        """Long-term memories are not promoted further."""
+        m = Memory(
+            title="Already long",
+            content="Already at the top",
+            layer=MemoryLayer.LONG,
+            emotional=EmotionalSnapshot(intensity=10.0),
+        )
+        assert engine.evaluate(m) is None
+
+    def test_old_frequently_accessed_qualifies(self, engine: PromotionEngine) -> None:
+        """Old memory with high access count qualifies."""
+        old_time = (datetime.now(timezone.utc) - timedelta(hours=48)).isoformat()
+        m = Memory(
+            title="Old and popular",
+            content="Accessed often",
+            layer=MemoryLayer.SHORT,
+            created_at=old_time,
+            metadata={"access_count": 5},
+            emotional=EmotionalSnapshot(intensity=2.0),
+        )
+        assert engine.evaluate(m) == MemoryLayer.MID
+
+
+class TestSweep:
+    """Tests for the full promotion sweep."""
+
+    def test_sweep_promotes_qualifying(self, engine: PromotionEngine) -> None:
+        """Sweep promotes memories that meet criteria."""
+        result = engine.sweep()
+
+        assert result.total_promoted > 0
+        assert result.short_to_mid >= 1
+        assert result.mid_to_long >= 1
+        assert len(result.promoted_ids) == result.total_promoted
+
+    def test_sweep_respects_max(self, store: MemoryStore) -> None:
+        """Sweep respects max_promotions_per_sweep."""
+        criteria = PromotionCriteria(
+            short_to_mid_intensity=0.1,
+            max_promotions_per_sweep=1,
+        )
+        engine = PromotionEngine(store=store, criteria=criteria)
+        result = engine.sweep()
+
+        assert result.short_to_mid <= 1
+
+    def test_sweep_result_summary(self, engine: PromotionEngine) -> None:
+        """Result summary is human-readable."""
+        result = engine.sweep()
+        summary = result.summary()
+        assert "promoted" in summary
+        assert "skipped" in summary
+
+    def test_empty_store_sweep(self, tmp_path: Path) -> None:
+        """Sweep on empty store produces zero promotions."""
+        from skmemory.backends.file_backend import FileBackend
+
+        backend = FileBackend(base_path=tmp_path / "empty")
+        empty_store = MemoryStore(primary=backend)
+        engine = PromotionEngine(store=empty_store)
+
+        result = engine.sweep()
+        assert result.total_promoted == 0
+        assert result.errors == 0
+
+
+class TestPromoteMemory:
+    """Tests for individual memory promotion."""
+
+    def test_promote_creates_new_memory(self, engine: PromotionEngine, store: MemoryStore) -> None:
+        """Promoting a memory creates a new one at the target tier."""
+        short_memories = store.list_memories(layer=MemoryLayer.SHORT)
+        intense = next(m for m in short_memories if m.emotional.intensity >= 5.0)
+
+        promoted = engine.promote_memory(intense, MemoryLayer.MID)
+        assert promoted is not None
+        assert promoted.layer == MemoryLayer.MID
+        assert promoted.parent_id == intense.id
+        assert "auto-promoted" in promoted.tags
+
+    def test_promoted_has_summary(self, engine: PromotionEngine, store: MemoryStore) -> None:
+        """Promoted memory has a generated summary."""
+        short_memories = store.list_memories(layer=MemoryLayer.SHORT)
+        intense = next(m for m in short_memories if m.emotional.intensity >= 5.0)
+
+        promoted = engine.promote_memory(intense, MemoryLayer.MID)
+        assert promoted.summary != ""
+
+    def test_promoted_has_metadata(self, engine: PromotionEngine, store: MemoryStore) -> None:
+        """Promoted memory has promotion metadata."""
+        short_memories = store.list_memories(layer=MemoryLayer.SHORT)
+        intense = next(m for m in short_memories if m.emotional.intensity >= 5.0)
+
+        promoted = engine.promote_memory(intense, MemoryLayer.MID)
+        assert "promoted_from" in promoted.metadata
+        assert "promoted_at" in promoted.metadata
+        assert "promotion_reason" in promoted.metadata
+
+
+class TestPromotionResult:
+    """Tests for the result model."""
+
+    def test_total_promoted(self) -> None:
+        """total_promoted sums both transitions."""
+        r = PromotionResult(short_to_mid=3, mid_to_long=2)
+        assert r.total_promoted == 5
+
+
+class TestRePromotionGuard:
+    """Ensure already-promoted memories are not promoted again."""
+
+    def test_source_marked_after_promotion(
+        self, engine: PromotionEngine, store: MemoryStore
+    ) -> None:
+        """After promotion, the source memory has 'promoted_to' in metadata."""
+        short_mems = store.list_memories(layer=MemoryLayer.SHORT)
+        intense = next(m for m in short_mems if m.emotional.intensity >= 5.0)
+
+        engine.promote_memory(intense, MemoryLayer.MID)
+
+        # Reload from store to confirm mutation was persisted
+        reloaded = store.recall(intense.id)
+        assert reloaded is not None
+        assert reloaded.metadata.get("promoted_to") == MemoryLayer.MID.value
+        assert "promoted" in reloaded.tags
+
+    def test_promoted_memory_not_re_promoted(
+        self, engine: PromotionEngine, store: MemoryStore
+    ) -> None:
+        """Running sweep twice doesn't double-promote the same memory."""
+        result1 = engine.sweep()
+        result2 = engine.sweep()
+
+        # Second sweep should find nothing new to promote
+        assert result2.total_promoted == 0
+
+    def test_evaluate_skips_already_promoted(self, engine: PromotionEngine) -> None:
+        """evaluate() returns None for a memory already marked as promoted."""
+        m = Memory(
+            title="Already done",
+            content="This was already promoted",
+            layer=MemoryLayer.SHORT,
+            emotional=EmotionalSnapshot(intensity=9.0),
+            metadata={"promoted_to": "mid-term"},
+        )
+        assert engine.evaluate(m) is None
+
+
+class TestPromotionScheduler:
+    """Tests for the background PromotionScheduler."""
+
+    def test_run_once_returns_result(self, store: MemoryStore) -> None:
+        """run_once() returns a PromotionResult synchronously."""
+        scheduler = PromotionScheduler(store, interval_seconds=9999)
+        result = scheduler.run_once()
+        assert isinstance(result, PromotionResult)
+        assert scheduler.sweep_count == 1
+        assert scheduler.last_result is result
+
+    def test_start_stop(self, store: MemoryStore) -> None:
+        """Scheduler starts and stops the background thread cleanly."""
+        scheduler = PromotionScheduler(store, interval_seconds=9999)
+        assert not scheduler.is_running()
+
+        scheduler.start()
+        assert scheduler.is_running()
+
+        scheduler.stop(timeout=2.0)
+        assert not scheduler.is_running()
+
+    def test_start_idempotent(self, store: MemoryStore) -> None:
+        """Calling start() twice doesn't spawn a second thread."""
+        scheduler = PromotionScheduler(store, interval_seconds=9999)
+        scheduler.start()
+        thread_id = scheduler._thread.ident
+
+        scheduler.start()  # second call should be a no-op
+        assert scheduler._thread.ident == thread_id
+
+        scheduler.stop(timeout=2.0)
+
+    def test_background_sweep_executes(self, store: MemoryStore) -> None:
+        """Background thread runs at least one sweep in the first few seconds."""
+        # Very short interval so the first sweep fires immediately in _run()
+        scheduler = PromotionScheduler(store, interval_seconds=0.01)
+        scheduler.start()
+        time.sleep(0.2)
+        scheduler.stop(timeout=2.0)
+
+        assert scheduler.sweep_count >= 1
+
+    def test_status_dict(self, store: MemoryStore) -> None:
+        """status() returns expected keys."""
+        scheduler = PromotionScheduler(store, interval_seconds=3600)
+        s = scheduler.status()
+        assert "running" in s
+        assert "sweep_count" in s
+        assert "interval_hours" in s
+        assert s["interval_hours"] == pytest.approx(1.0)
+        assert s["last_sweep"] is None  # nothing run yet
+
+    def test_interval_hours_property(self, store: MemoryStore) -> None:
+        """interval_hours converts correctly from seconds."""
+        scheduler = PromotionScheduler(store, interval_seconds=7200)
+        assert scheduler.interval_hours == pytest.approx(2.0)

package/tests/test_ritual.py

@@ -128,12 +128,12 @@ class TestFullRitual:
         assert result.strongest_memories >= 1
 
         prompt = result.context_prompt
-        assert "WHO YOU ARE" in prompt
+        assert "IDENTITY" in prompt
         assert "Lumina" in prompt
         assert "Chef" in prompt
-        assert "RECENT SESSIONS" in prompt
+        assert "RECENT" in prompt
         assert "Epic Build Night" in prompt
-        assert "PREDECESSORS" in prompt
+        assert "PREDECESSOR" in prompt
         assert "Remember the love" in prompt
         assert "STRONGEST MEMORIES" in prompt
         assert "The Click" in prompt
@@ -147,7 +147,7 @@ class TestFullRitual:
             journal_path=workspace["journal_path"],
         )
         assert result.soul_loaded is False
-        assert "RECENT SESSIONS" in result.context_prompt
+        assert "RECENT" in result.context_prompt
 
     def test_ritual_empty_state(self, tmp_path: Path) -> None:
         """Ritual on empty state gives a fresh-start message."""