feat(compression): flip in_place default to True + guard fix (#38763) [salvage #51959]

gateway/run.py

@@ -9578,7 +9578,7 @@ async def _handle_message_with_agent(self, event, source, _quick_key: str, run_g
                                     _hyg_new_sid = _hyg_agent.session_id
                                     _hyg_rotated = _hyg_new_sid != session_entry.session_id
                                     _hyg_in_place = bool(
-                                        getattr(_hyg_agent, "compression_in_place", False)
+                                        getattr(_hyg_agent, "_last_compaction_in_place", False)
                                     )
                                     if _hyg_rotated:
                                         session_entry.session_id = _hyg_new_sid

gateway/slash_commands.py

@@ -2859,7 +2859,7 @@ async def _handle_compress_command(self, event: MessageEvent) -> str:
                 # transcript replaced with the compacted set).
                 new_session_id = tmp_agent.session_id
                 rotated = new_session_id != session_entry.session_id
-                _in_place = bool(getattr(tmp_agent, "compression_in_place", False))
+                _in_place = bool(getattr(tmp_agent, "_last_compaction_in_place", False))
                 if rotated:
                     session_entry.session_id = new_session_id
                     self.session_store._save()

hermes_cli/config.py

@@ -1315,7 +1315,7 @@ def _ensure_hermes_home_managed(home: Path):
                                       # exact route is affected — gpt-5.5 on OpenAI's
                                       # direct API, OpenRouter, and Copilot keep the
                                       # global threshold regardless.
-        "in_place": False,            # When True, compaction rewrites the message
+        "in_place": True,             # When True, compaction rewrites the message
                                       # list and rebuilds the system prompt WITHOUT
                                       # rotating the session id — the conversation
                                       # keeps one durable id for its whole life

scripts/release.py

@@ -78,6 +78,7 @@
     "87440198+JoaoMarcos44@users.noreply.github.com": "JoaoMarcos44",
     "joaomarcosdias444@gmail.com": "JoaoMarcos44",
     "286497132+srojk34@users.noreply.github.com": "srojk34",
+    "srojk34@users.noreply.github.com": "srojk34",  # legacy prefix-less noreply (PR #50098 salvage; #38763)
     "59806492+sitkarev@users.noreply.github.com": "sitkarev",
     "zheng@omegasys.eu": "omegazheng",
     "220877172+james47kjv@users.noreply.github.com": "james47kjv",

tests/agent/test_compression_concurrent_fork.py

@@ -77,6 +77,10 @@ def _compress_with_overlap(*_a, **_kw):
     compressor._last_aux_model_failure_model = None
     compressor._last_aux_model_failure_error = None
     agent.context_compressor = compressor
+    # These tests cover the ROTATION fallback path (forking, child sessions,
+    # lock contention) — pin in_place=False so they keep exercising it
+    # regardless of the global default (which flipped to True in #38763).
+    agent.compression_in_place = False
     return agent
 
 

tests/agent/test_compression_logging_session_context.py

@@ -50,6 +50,10 @@ def _build_agent_with_db(db: SessionDB, session_id: str):
     compressor._last_aux_model_failure_model = None
     compressor._last_aux_model_failure_error = None
     agent.context_compressor = compressor
+    # This test covers the ROTATION fallback (logging session-context follows
+    # the id rotation) — pin in_place=False so it keeps exercising rotation
+    # regardless of the global default (flipped to True in #38763).
+    agent.compression_in_place = False
     return agent
 
 

tests/agent/test_compression_rotation_state.py

@@ -54,6 +54,9 @@ def _build_agent_with_db(db: SessionDB, session_id: str, platform: str = "telegr
     compressor._last_aux_model_failure_model = None
     compressor._last_aux_model_failure_error = None
     agent.context_compressor = compressor
+    # ROTATION fallback path — pin in_place=False so these keep covering fork
+    # rotation regardless of the global default (flipped to True in #38763).
+    agent.compression_in_place = False
     return agent
 
 

tests/gateway/test_compression_concurrent_sessions.py

@@ -71,6 +71,10 @@ def _compress_with_overlap(*_a, **_kw):
     compressor._last_aux_model_failure_model = None
     compressor._last_aux_model_failure_error = None
     agent.context_compressor = compressor
+    # ROTATION fallback path — pin in_place=False so these keep covering the
+    # concurrent-rotation lock contract regardless of the global default
+    # (flipped to True in #38763).
+    agent.compression_in_place = False
     return agent
 
 

tests/gateway/test_session_hygiene.py

@@ -494,6 +494,105 @@ def _compress_context(self, messages, *_args, **_kwargs):
     runner.session_store.rewrite_transcript.assert_not_called()
 
 
+@pytest.mark.asyncio
+async def test_session_hygiene_preserves_transcript_when_in_place_configured_but_no_db(monkeypatch, tmp_path):
+    """Regression: when compression.in_place is True but the hygiene agent has
+    no session_db, archive_and_compact cannot run — _last_compaction_in_place
+    stays False.  The guard must read the *result* flag, not the *config* flag,
+    otherwise the transcript is unconditionally rewritten with only the summary
+    (permanent data loss identical to #21301)."""
+    fake_dotenv = types.ModuleType("dotenv")
+    fake_dotenv.load_dotenv = lambda *args, **kwargs: None
+    monkeypatch.setitem(sys.modules, "dotenv", fake_dotenv)
+
+    class InPlaceConfiguredAgent:
+        last_instance = None
+
+        def __init__(self, **kwargs):
+            self.model = kwargs.get("model")
+            self.session_id = kwargs.get("session_id", "fake-session")
+            self.compression_in_place = True
+            self._last_compaction_in_place = False
+            self._print_fn = None
+            self.shutdown_memory_provider = MagicMock()
+            self.close = MagicMock()
+            type(self).last_instance = self
+
+        def _compress_context(self, messages, *_args, **_kwargs):
+            return ([{"role": "assistant", "content": "summary only"}], None)
+
+    fake_run_agent = types.ModuleType("run_agent")
+    fake_run_agent.AIAgent = InPlaceConfiguredAgent
+    monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent)
+
+    gateway_run = importlib.import_module("gateway.run")
+    GatewayRunner = gateway_run.GatewayRunner
+
+    adapter = HygieneCaptureAdapter()
+    runner = object.__new__(GatewayRunner)
+    runner.config = GatewayConfig(
+        platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="fake-token")}
+    )
+    runner.adapters = {Platform.TELEGRAM: adapter}
+    runner._voice_mode = {}
+    runner.hooks = SimpleNamespace(emit=AsyncMock(), loaded_hooks=False)
+    runner.session_store = MagicMock()
+    runner.session_store.get_or_create_session.return_value = SessionEntry(
+        session_key="agent:main:telegram:group:-1001:17585",
+        session_id="sess-1",
+        created_at=datetime.now(),
+        updated_at=datetime.now(),
+        platform=Platform.TELEGRAM,
+        chat_type="group",
+    )
+    runner.session_store.load_transcript.return_value = _make_history(6, content_size=400)
+    runner.session_store.has_any_sessions.return_value = True
+    runner.session_store.rewrite_transcript = MagicMock()
+    runner.session_store.append_to_transcript = MagicMock()
+    runner._running_agents = {}
+    runner._pending_messages = {}
+    runner._pending_approvals = {}
+    runner._session_db = None
+    runner._is_user_authorized = lambda _source: True
+    runner._set_session_env = lambda _context: None
+    runner._run_agent = AsyncMock(
+        return_value={
+            "final_response": "ok",
+            "messages": [],
+            "tools": [],
+            "history_offset": 0,
+            "last_prompt_tokens": 0,
+        }
+    )
+
+    monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path)
+    monkeypatch.setattr(gateway_run, "_resolve_runtime_agent_kwargs", lambda: {"api_key": "fake"})
+    monkeypatch.setattr(
+        "agent.model_metadata.get_model_context_length",
+        lambda *_args, **_kwargs: 100,
+    )
+    monkeypatch.setenv("TELEGRAM_HOME_CHANNEL", "795544298")
+
+    event = MessageEvent(
+        text="hello",
+        source=SessionSource(
+            platform=Platform.TELEGRAM,
+            chat_id="-1001",
+            chat_type="group",
+            thread_id="17585",
+            user_id="12345",
+        ),
+        message_id="1",
+    )
+
+    result = await runner._handle_message(event)
+
+    assert result == "ok"
+    # The config says in_place=True, but the DB write failed (no session_db)
+    # so _last_compaction_in_place is False. Transcript must NOT be rewritten.
+    runner.session_store.rewrite_transcript.assert_not_called()
+
+
 @pytest.mark.asyncio
 async def test_session_hygiene_warns_user_when_compression_aborts(monkeypatch, tmp_path):
     """When auxiliary compression's summary LLM call fails, the compressor

tests/run_agent/test_compression_boundary_hook.py

@@ -22,7 +22,7 @@ class TestCompressionBoundaryHook:
     def _make_agent(self, session_db):
         with patch.dict(os.environ, {"OPENROUTER_API_KEY": "test-key"}):
             from run_agent import AIAgent
-            return AIAgent(
+            agent = AIAgent(
                 api_key="test-key",
                 base_url="https://openrouter.ai/api/v1",
                 model="test/model",
@@ -32,6 +32,9 @@ def _make_agent(self, session_db):
                 skip_context_files=True,
                 skip_memory=True,
             )
+            # ROTATION fallback — pin in_place=False regardless of default (#38763).
+            agent.compression_in_place = False
+            return agent
 
     def test_on_session_start_called_with_compression_boundary(self):
         from hermes_state import SessionDB
@@ -167,7 +170,7 @@ class TestSessionCompressEvent:
     def _make_agent(self, session_db, event_callback=None):
         with patch.dict(os.environ, {"OPENROUTER_API_KEY": "test-key"}):
             from run_agent import AIAgent
-            return AIAgent(
+            agent = AIAgent(
                 api_key="test-key",
                 base_url="https://openrouter.ai/api/v1",
                 model="test/model",
@@ -178,6 +181,9 @@ def _make_agent(self, session_db, event_callback=None):
                 skip_memory=True,
                 event_callback=event_callback,
             )
+            # ROTATION fallback — pin in_place=False regardless of default (#38763).
+            agent.compression_in_place = False
+            return agent
 
     def _stub_compressor(self):
         compressor = MagicMock()

tests/run_agent/test_in_place_compaction.py

@@ -184,9 +184,11 @@ def test_rotation_still_preflushes(self):
             assert calls["n"] == 1
 
 
-class TestRotationStillDefault:
+class TestRotationFallbackWhenFlagOff:
     def test_rotation_when_flag_off(self):
-        """Regression guard: flag off => legacy rotation is unchanged."""
+        """Rotation is now the OPT-OUT fallback (default flipped to in-place in
+        #38763). With in_place=False explicitly set, legacy rotation is
+        unchanged — forks a renamed continuation session."""
         from hermes_state import SessionDB
         from agent.conversation_compression import compress_context
 
@@ -247,10 +249,12 @@ def test_signal_set_on_in_place_unset_on_rotation(self):
 
 
 class TestInPlaceConfigDefault:
-    def test_flag_defaults_off(self):
+    def test_flag_defaults_on(self):
+        """In-place is the default as of #38763 (rotation is now opt-out via
+        compression.in_place: false)."""
         from hermes_cli.config import DEFAULT_CONFIG
 
-        assert DEFAULT_CONFIG["compression"].get("in_place") is False
+        assert DEFAULT_CONFIG["compression"].get("in_place") is True
 
 
 class TestCompactedTurnsStaySearchable: