fix(honcho): preserve delayed and rewritten recall context

plugins/memory/honcho/__init__.py

@@ -754,9 +754,12 @@ class HonchoMemoryProvider(MemoryProvider):
 
                 def _fetch_base() -> None:
                     try:
-                        _ctx_holder["ctx"] = self._manager.get_prefetch_context(
+                        ctx = self._manager.get_prefetch_context(
                             self._session_key, query or None
                         ) or {}
+                        _ctx_holder["ctx"] = ctx
+                        if ctx:
+                            self._manager.set_context_result(self._session_key, ctx)
                     except Exception as e:
                         logger.debug("Honcho first-turn base context failed: %s", e)
 
@@ -773,6 +776,7 @@ class HonchoMemoryProvider(MemoryProvider):
                 _bt.join(timeout=_base_wait)
                 _ctx = _ctx_holder.get("ctx")
                 if _ctx:
+                    self._manager.pop_context_result(self._session_key)
                     formatted = self._format_first_turn_context(_ctx)
                     if formatted:
                         with self._base_context_lock:
@@ -1248,7 +1252,9 @@ class HonchoMemoryProvider(MemoryProvider):
                     # nothing.
                     logger.debug("Honcho dialectic depth %d: pass %d has no non-empty prior — "
                                  "falling back to base prompt", self._dialectic_depth, i)
-                    prompt = self._build_dialectic_prompt(0, prior_results, is_cold)
+                    prompt = rewritten_query or self._build_dialectic_prompt(
+                        0, prior_results, is_cold
+                    )
                 else:
                     prompt = self._build_dialectic_prompt(i, prior_results, is_cold)
 

tests/honcho_plugin/test_query_rewrite.py

@@ -147,6 +147,17 @@ def test_query_rewriter_runs_once_for_a_multi_pass_dialectic_cycle():
     assert provider._manager.dialectic_query.call_count == 2
 
 
+def test_empty_first_pass_retries_with_rewritten_query():
+    rewritten = "What prior deployment decisions did the user make?"
+    provider = _provider(lambda message: rewritten, depth=2)
+    provider._manager.dialectic_query.side_effect = ["", "grounded synthesis"]
+
+    provider._run_dialectic_depth("What should we deploy?")
+
+    prompts = [call.args[1] for call in provider._manager.dialectic_query.call_args_list]
+    assert prompts == [rewritten, rewritten]
+
+
 def test_session_prewarm_can_skip_query_rewrite():
     rewriter = MagicMock(return_value="unused")
     provider = _provider(rewriter)

tests/honcho_plugin/test_session.py

@@ -1115,6 +1115,45 @@ class TestBaseContextSummary:
         formatted = provider._format_first_turn_context(ctx)
         assert "Session Summary" not in formatted
 
+    def test_timed_out_first_turn_context_surfaces_next_turn(self):
+        import threading
+        import time
+
+        ready = threading.Event()
+        cached = {}
+        manager = MagicMock()
+
+        def get_context(*args, **kwargs):
+            ready.wait(timeout=1)
+            return {"representation": "late user context", "card": ""}
+
+        manager.get_prefetch_context.side_effect = get_context
+        manager.set_context_result.side_effect = (
+            lambda session_key, result: cached.__setitem__(session_key, result)
+        )
+        manager.pop_context_result.side_effect = (
+            lambda session_key: cached.pop(session_key, {})
+        )
+
+        provider = HonchoMemoryProvider()
+        provider._manager = manager
+        provider._config = SimpleNamespace(timeout=0.01, context_tokens=0)
+        provider._session_key = "test"
+        provider._session_initialized = True
+        provider._recall_mode = "context"
+        provider._turn_count = 1
+        provider._last_dialectic_turn = 0
+
+        assert provider.prefetch("first question") == ""
+        ready.set()
+
+        deadline = time.monotonic() + 1
+        while "test" not in cached and time.monotonic() < deadline:
+            time.sleep(0.01)
+
+        provider._turn_count = 2
+        assert "late user context" in provider.prefetch("follow-up question")
+
 
 class TestDialecticDepth:
     """Tests for the dialecticDepth multi-pass system."""