From 164d024bb3356adc150fa339e7628a13e1a28270 Mon Sep 17 00:00:00 2001
From: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
Date: Tue, 6 Oct 2026 09:48:31 +0530
Subject: [PATCH] test(compaction): move the bench probe rebuild test into its
 own file

Code-health ratchet: tests/agent/test_context_compressor.py grew to
3847 > 3829. New tests go in a topical test_<stem>_<topic>.py.
---
 tests/agent/test_context_compressor.py        | 24 -----------
 .../test_context_compressor_bench_probe.py    | 43 +++++++++++++++++++
 2 files changed, 43 insertions(+), 24 deletions(-)
 create mode 100644 tests/agent/test_context_compressor_bench_probe.py

diff --git a/tests/agent/test_context_compressor.py b/tests/agent/test_context_compressor.py
index a4c8af3984..87b854154d 100644
--- a/tests/agent/test_context_compressor.py
+++ b/tests/agent/test_context_compressor.py
@@ -3685,30 +3685,6 @@ class TestPreLlmFeasibilityCheck:
         mock_gen.assert_called_once()
         assert compressor._prellm_skip_count == 0
 
-    def test_bench_probe_deadline_survives_a_compressor_rebuild(self, compressor, tmp_path):
-        """The gateway rebuilds the agent on cache eviction: a fresh compressor bound to the same
-        session must resume the persisted bench window, not restart it (else the probe starves)."""
-        db = SessionDB(db_path=tmp_path / "state.db")
-        db.create_session("s1", "cli")
-        db.set_compression_fallback_streak("s1", 2)
-        compressor.bind_session_state(db, "s1")
-        window = compressor._ANTI_THRASH_RECOVERY_SECONDS
-
-        with patch("agent.context_compressor.time.time", return_value=1000.0), \
-                patch.object(compressor, "_generate_summary", return_value="LLM summary") as mock_gen:
-            compressor.compress(self._make_messages(), force=False)  # benched, arms the window
-        mock_gen.assert_not_called()
-
-        with patch("agent.context_compressor.get_model_context_length", return_value=100000):
-            rebuilt = ContextCompressor(
-                model="test/model", threshold_percent=0.85, protect_first_n=2, protect_last_n=2, quiet_mode=True,
-            )
-        rebuilt.bind_session_state(db, "s1")
-        with patch("agent.context_compressor.time.time", return_value=1000.0 + window + 1), \
-                patch.object(rebuilt, "_generate_summary", return_value="LLM summary") as mock_gen:
-            rebuilt.compress(self._make_messages(), force=False)
-        mock_gen.assert_called_once()
-
     def test_skip_fires_on_fat_tail_small_middle(self, compressor):
         """The target scenario from #60451: a tool-heavy transcript whose
         protected tail already holds most of the tokens, leaving a tiny
diff --git a/tests/agent/test_context_compressor_bench_probe.py b/tests/agent/test_context_compressor_bench_probe.py
new file mode 100644
index 0000000000..f9227397e2
--- /dev/null
+++ b/tests/agent/test_context_compressor_bench_probe.py
@@ -0,0 +1,43 @@
+"""The summary-model bench (fallback streak >= 2) and its persisted probe window."""
+
+from unittest.mock import patch
+
+from agent.context_compressor import ContextCompressor
+from hermes_state import SessionDB
+
+
+def _compressor():
+    with patch("agent.context_compressor.get_model_context_length", return_value=100000):
+        c = ContextCompressor(model="test/model", threshold_percent=0.85, protect_first_n=2, protect_last_n=2, quiet_mode=True)
+        _ = c.context_length
+        return c
+
+
+def _messages(n_pairs=10):
+    msgs = [{"role": "system", "content": "system prompt"}]
+    for i in range(n_pairs):
+        msgs += [{"role": "user", "content": f"question {i}"}, {"role": "assistant", "content": "short reply"}]
+    return msgs
+
+
+def test_bench_probe_deadline_survives_a_compressor_rebuild(tmp_path):
+    """The gateway rebuilds the agent on cache eviction: a fresh compressor bound to the same
+    session must resume the persisted bench window, not restart it (else the probe starves)."""
+    db = SessionDB(db_path=tmp_path / "state.db")
+    db.create_session("s1", "cli")
+    db.set_compression_fallback_streak("s1", 2)
+    compressor = _compressor()
+    compressor.bind_session_state(db, "s1")
+    window = compressor._ANTI_THRASH_RECOVERY_SECONDS
+
+    with patch("agent.context_compressor.time.time", return_value=1000.0), \
+            patch.object(compressor, "_generate_summary", return_value="LLM summary") as mock_gen:
+        compressor.compress(_messages(), force=False)  # benched, arms the window
+    mock_gen.assert_not_called()
+
+    rebuilt = _compressor()
+    rebuilt.bind_session_state(db, "s1")
+    with patch("agent.context_compressor.time.time", return_value=1000.0 + window + 1), \
+            patch.object(rebuilt, "_generate_summary", return_value="LLM summary") as mock_gen:
+        rebuilt.compress(_messages(), force=False)
+    mock_gen.assert_called_once()