feat: add memory management pipeline

Admission control, entropy-based micro-faulting, phantom tool
injection for backing store queries, and xMemory session hierarchy
for long conversations (50+ turns).

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

tests/test_admission.py

@@ -0,0 +1,156 @@
+"""Tests for the admission control module.
+
+Tests the AdmissionController scoring logic across all four axes:
+type_score, novelty_score, utility_score, size_score, and the
+threshold-based admit/reject decision.
+"""
+
+from __future__ import annotations
+
+from mnemosyne.admission import AdmissionController, AdmissionScore, DEFAULT_THRESHOLD
+
+
+# ── Type score ───────────────────────────────────────────────────────────
+
+
+def test_admission_score_design_decision():
+    """design_decision has the highest type weight (0.9)."""
+    ctrl = AdmissionController()
+    score = ctrl.score("x" * 300, "design_decision", has_duplicate=False)
+    assert score.type_score == 0.9
+
+
+def test_admission_score_conversation_phase():
+    """conversation_phase has the lowest type weight (0.3)."""
+    ctrl = AdmissionController()
+    score = ctrl.score("x" * 300, "conversation_phase", has_duplicate=False)
+    assert score.type_score == 0.3
+
+
+def test_admission_unknown_type_default():
+    """Unknown object types default to 0.3."""
+    ctrl = AdmissionController()
+    score = ctrl.score("x" * 300, "totally_unknown_type", has_duplicate=False)
+    assert score.type_score == 0.3
+
+
+# ── Novelty score ────────────────────────────────────────────────────────
+
+
+def test_admission_duplicate_penalized():
+    """Duplicates get novelty_score=0.2 instead of 1.0."""
+    ctrl = AdmissionController()
+    score_unique = ctrl.score("x" * 300, "file_context", has_duplicate=False)
+    score_dup = ctrl.score("x" * 300, "file_context", has_duplicate=True)
+    assert score_unique.novelty_score == 1.0
+    assert score_dup.novelty_score == 0.2
+    assert score_dup.total < score_unique.total
+
+
+# ── Utility score ────────────────────────────────────────────────────────
+
+
+def test_admission_utility_code_blocks():
+    """Content with code blocks gets utility boost."""
+    ctrl = AdmissionController()
+    content_with_code = "Here is code:\n```python\ndef foo(): pass\n```\n" + "x" * 200
+    content_plain = "Just some plain text without code blocks. " + "x" * 200
+    score_code = ctrl.score(content_with_code, "file_context", has_duplicate=False)
+    score_plain = ctrl.score(content_plain, "file_context", has_duplicate=False)
+    assert score_code.utility_score > score_plain.utility_score
+
+
+def test_admission_utility_entities():
+    """Content with many key_entities gets utility boost."""
+    ctrl = AdmissionController()
+    entities = ["foo.py", "bar.py", "baz.py"]
+    score_with = ctrl.score("x" * 300, "file_context", has_duplicate=False, key_entities=entities)
+    score_without = ctrl.score("x" * 300, "file_context", has_duplicate=False, key_entities=[])
+    assert score_with.utility_score > score_without.utility_score
+
+
+# ── Size score ───────────────────────────────────────────────────────────
+
+
+def test_admission_size_small_penalized():
+    """Very small content (< 100 chars) gets penalized size_score."""
+    ctrl = AdmissionController()
+    score = ctrl.score("tiny", "file_context", has_duplicate=False)
+    assert score.size_score < 1.0
+    assert score.size_score == len("tiny") / 100
+
+
+def test_admission_size_large_penalized():
+    """Very large content (> 50k chars) gets penalized size_score."""
+    ctrl = AdmissionController()
+    score = ctrl.score("x" * 80000, "file_context", has_duplicate=False)
+    assert score.size_score < 1.0
+    assert score.size_score >= 0.3
+
+
+def test_admission_size_normal():
+    """Normal-sized content (100-50k chars) gets size_score=1.0."""
+    ctrl = AdmissionController()
+    score = ctrl.score("x" * 500, "file_context", has_duplicate=False)
+    assert score.size_score == 1.0
+
+
+# ── Threshold decisions ──────────────────────────────────────────────────
+
+
+def test_admission_threshold_admit():
+    """High-value content is admitted."""
+    ctrl = AdmissionController()
+    # design_decision (0.9 type) + unique (1.0 novelty) + code + entities + length
+    content = "```python\ndef important(): pass\n```\n" + "x" * 300
+    admitted, score = ctrl.should_admit(
+        content,
+        "design_decision",
+        has_duplicate=False,
+        key_entities=["foo.py", "bar.py", "baz.py"],
+    )
+    assert admitted is True
+    assert score.total >= DEFAULT_THRESHOLD
+
+
+def test_admission_threshold_reject():
+    """Low-value duplicate conversation_phase is rejected."""
+    ctrl = AdmissionController()
+    # conversation_phase (0.3 type) + duplicate (0.2 novelty) + tiny content
+    admitted, score = ctrl.should_admit("hi", "conversation_phase", has_duplicate=True)
+    assert admitted is False
+    assert score.total < DEFAULT_THRESHOLD
+
+
+# ── Stats tracking ───────────────────────────────────────────────────────
+
+
+def test_admission_stats_tracking():
+    """Stats correctly count admitted and rejected objects."""
+    ctrl = AdmissionController()
+    assert ctrl.stats == {"admitted": 0, "rejected": 0}
+
+    # Admit a high-value object
+    ctrl.should_admit(
+        "```python\ndef foo(): pass\n```\n" + "x" * 300,
+        "design_decision",
+        has_duplicate=False,
+        key_entities=["a.py", "b.py", "c.py"],
+    )
+    assert ctrl.stats["admitted"] == 1
+
+    # Reject a low-value object
+    ctrl.should_admit("hi", "conversation_phase", has_duplicate=True)
+    assert ctrl.stats["rejected"] == 1
+    assert ctrl.stats["admitted"] == 1
+
+
+# ── Edge cases ───────────────────────────────────────────────────────────
+
+
+def test_admission_empty_content():
+    """Empty content gets size_score=0 and low utility."""
+    ctrl = AdmissionController()
+    score = ctrl.score("", "file_context", has_duplicate=False)
+    assert score.size_score == 0.0
+    assert score.utility_score == 0.0