fix: wait for agent_case LanceDB visibility before skill extraction

src/everos/memory/strategies/extract_agent_skill.py

@@ -34,6 +34,8 @@
 
 from __future__ import annotations
 
+import asyncio
+
 from everalgo.agent_memory import AgentSkillExtractor
 from everalgo.types import AgentCase as AlgoAgentCase
 from everalgo.types import AgentSkill as AlgoAgentSkill
@@ -87,6 +89,15 @@
 Ranking is ``(quality_score desc, timestamp desc)`` — same ordering
 opensource ``AgentSkillExtractor._load_case_history`` applies."""
 
+CASE_VISIBILITY_ATTEMPTS = 6
+CASE_VISIBILITY_POLL_SECONDS = 0.2
+"""Bounded wait for cascade to surface the fresh AgentCase in LanceDB.
+
+``SkillClusterUpdated`` can arrive before the md→LanceDB write has
+become visible. A short local poll absorbs the normal watcher/indexer
+lag without consuming the whole OME retry budget in a tight loop.
+"""
+
 
 class _ClusterMissingError(RuntimeError):
     """Race with the cluster strategy; OME retry will catch up."""
@@ -198,16 +209,21 @@ async def _load_target_case(
     app_id: str,
     project_id: str,
 ) -> LanceAgentCase:
-    """Pull the target case row, raising a retry-class error on cascade lag."""
-    target = await agent_case_repo.find_by_owner_entry(
-        agent_id, case_entry_id, app_id=app_id, project_id=project_id
-    )
-    if target is None:
-        # Cascade hasn't indexed the freshly-written md yet.
-        raise _CaseNotYetIndexedError(
-            f"AgentCase entry_id={case_entry_id} not in LanceDB yet; retrying"
+    """Pull the target case row, allowing a short wait for cascade lag."""
+    for attempt in range(CASE_VISIBILITY_ATTEMPTS):
+        target = await agent_case_repo.find_by_owner_entry(
+            agent_id, case_entry_id, app_id=app_id, project_id=project_id
         )
-    return target
+        if target is not None:
+            return target
+        if attempt < CASE_VISIBILITY_ATTEMPTS - 1:
+            await asyncio.sleep(CASE_VISIBILITY_POLL_SECONDS)
+
+    raise _CaseNotYetIndexedError(
+        "AgentCase entry_id="
+        f"{case_entry_id} not in LanceDB after "
+        f"{CASE_VISIBILITY_ATTEMPTS} visibility checks; retrying"
+    )
 
 
 async def _select_existing_skills(

tests/unit/test_memory/test_strategies/test_extract_agent_skill.py

@@ -38,6 +38,7 @@
     _CaseNotYetIndexedError,
     _ClusterMissingError,
     _collect_supporting_entry_ids,
+    _load_target_case,
     _resolve_query_vector,
     _select_existing_skills,
     _select_supporting_cases,
@@ -161,13 +162,40 @@ async def test_raises_when_target_case_not_yet_in_lancedb() -> None:
         patch(
             "everos.memory.strategies.extract_agent_skill.agent_case_repo"
         ) as mock_case_repo,
+        patch("asyncio.sleep", new_callable=AsyncMock),
     ):
         mock_cluster_repo.get_with_members = AsyncMock(return_value=_algo_cluster())
         mock_case_repo.find_by_owner_entry = AsyncMock(return_value=None)
         with pytest.raises(_CaseNotYetIndexedError):
             await extract_agent_skill(_event(), FakeStrategyContext())
 
 
+async def test_load_target_case_waits_for_lancedb_visibility() -> None:
+    """Short cascade lag should not force an immediate retry/dead-letter."""
+    target = _lance_case("ac_20260517_0001")
+
+    with (
+        patch(
+            "everos.memory.strategies.extract_agent_skill.agent_case_repo"
+        ) as mock_case_repo,
+        patch("asyncio.sleep", new_callable=AsyncMock) as mock_sleep,
+    ):
+        mock_case_repo.find_by_owner_entry = AsyncMock(
+            side_effect=[None, None, target]
+        )
+
+        got = await _load_target_case(
+            "agent_42",
+            "ac_20260517_0001",
+            app_id="default",
+            project_id="default",
+        )
+
+    assert got is target
+    assert mock_case_repo.find_by_owner_entry.await_count == 3
+    assert mock_sleep.await_count == 2
+
+
 # ── end-to-end orchestration (mocked) ────────────────────────────────────