fix(context): bound session recall payloads

contributors/emails/joaomarcosdias444@gmail.com

@@ -0,0 +1,2 @@
+JoaoMarcos44
+# PR #55640 contributor attribution

tests/tools/test_session_search.py

@@ -12,6 +12,7 @@
 
 import pytest
 
+from agent.context_compressor import SUMMARY_PREFIX
 from hermes_state import SessionDB
 from tools.session_search_tool import (
     SESSION_SEARCH_SCHEMA,
@@ -152,6 +153,45 @@ def test_discovery_result_has_bookends_and_window(self, db):
             assert "messages_after" in hit
 
 
+    def test_title_match_filters_structured_summary_and_caps_content(self, db):
+        db.create_session("s_bounded_title", source="cli")
+        db.set_session_title("s_bounded_title", "bounded-title-only")
+        db.append_message(
+            "s_bounded_title",
+            role="user",
+            content=[{"type": "text", "text": "ordinary opening " + "o" * 5_000}],
+        )
+        db.append_message(
+            "s_bounded_title",
+            role="assistant",
+            content={"parts": [{"text": "provider wrapper " + "d" * 50_000}]},
+        )
+        db.append_message(
+            "s_bounded_title",
+            role="assistant",
+            content=[{"type": "text", "text": SUMMARY_PREFIX + " " + "s" * 50_000}],
+        )
+
+        result = json.loads(session_search(query="bounded-title-only", db=db))
+
+        assert result["success"] is True
+        assert result["count"] == 1
+        hit = result["results"][0]
+        assert hit["matched_role"] == "session_title"
+        bookends = hit["bookend_start"] + hit["bookend_end"]
+        assert all("[CONTEXT COMPACTION" not in msg.get("content", "") for msg in bookends)
+        assert all(len(msg.get("content", "")) <= 1_201 for msg in bookends)
+        assert any(
+            msg.get("content_truncated") and msg.get("original_content_chars", 0) > 1_200
+            for msg in bookends
+        )
+        assert all(len(msg.get("content", "")) <= 4_001 for msg in hit["messages"])
+        assert any(
+            msg.get("content_truncated") and msg.get("original_content_chars", 0) > 4_000
+            for msg in hit["messages"]
+        )
+
+
     def test_current_session_filtered_out(self, db):
         _seed_modpack_sessions(db)
         result = json.loads(session_search(query="modpack", db=db, current_session_id="s_newest"))
@@ -468,8 +508,12 @@ class TestCompactionSummaryFiltering:
 
     def test_is_compaction_summary_detects_prefix(self):
         from tools.session_search_tool import _is_compaction_summary
-        assert _is_compaction_summary("[CONTEXT COMPACTION — REFERENCE ONLY] foo")
+        assert _is_compaction_summary(SUMMARY_PREFIX + " generated summary")
         assert _is_compaction_summary("[CONTEXT SUMMARY]: old summary")
+        assert _is_compaction_summary(
+            [{"type": "text", "text": SUMMARY_PREFIX + " structured summary"}]
+        )
+        assert not _is_compaction_summary("[CONTEXT COMPACTION lookalike from a user")
         assert not _is_compaction_summary("Hello, how can I help?")
         assert not _is_compaction_summary("")
         assert not _is_compaction_summary(None)
@@ -478,9 +522,11 @@ def test_compaction_summary_excluded_from_bookend_start(self, db):
         """Compaction handoff in bookend_start position must be filtered out."""
         db.create_session("s_compact", source="cli")
         # First message: a compaction handoff (should be filtered)
-        db.append_message("s_compact", role="user",
-                          content="[CONTEXT COMPACTION — REFERENCE ONLY] "
-                                  "Earlier turns were compacted into the summary below. " + "x" * 50000)
+        db.append_message(
+            "s_compact",
+            role="user",
+            content=SUMMARY_PREFIX + " " + "x" * 50000,
+        )
         # Second message: normal user message
         db.append_message("s_compact", role="user", content="Fix the zorgblat rendering bug")
         # Padding messages to push window away from session start (so bookend has room)
@@ -685,9 +731,7 @@ def _seed_compacted_session(self, db):
         # Compact in place: everything above becomes active=0/compacted=1 and
         # the handoff summary is inserted as the new live tail.
         db.archive_and_compact("s_both", [
-            {"role": "user",
-             "content": "[CONTEXT COMPACTION — REFERENCE ONLY] "
-                        "Earlier turns were compacted into this summary. " + "s" * 50000},
+            {"role": "user", "content": SUMMARY_PREFIX + " " + "s" * 50000},
             {"role": "assistant", "content": "Continuing after compaction."},
         ])
         db._conn.commit()

tools/session_search_tool.py

@@ -67,16 +67,6 @@
     "session_started",
 )
 
-# Prefixes that identify generated context-compaction handoff summaries.
-# These are inserted by agent/context_compressor.py as normal user/assistant
-# messages but contain machine-generated summary metadata — not user content.
-# They must be excluded from discovery bookends to avoid re-introducing huge
-# compaction payloads into fresh sessions via session_search.  (#43175)
-_COMPACTION_PREFIXES = (
-    "[CONTEXT COMPACTION",
-    "[CONTEXT SUMMARY]:",
-)
-
 
 def _format_timestamp(ts: Union[int, float, str, None]) -> str:
     """Convert a Unix timestamp (float/int) or ISO string to a human-readable date.
@@ -103,12 +93,19 @@ def _format_timestamp(ts: Union[int, float, str, None]) -> str:
     return str(ts)
 
 
-def _is_compaction_summary(content: str) -> bool:
-    """Return True if *content* looks like a generated compaction handoff."""
-    if not content:
-        return False
-    stripped = content.lstrip()
-    return any(stripped.startswith(p) for p in _COMPACTION_PREFIXES)
+def _is_compaction_summary(content: Any) -> bool:
+    """Return True for a standalone compaction handoff in any supported shape."""
+    from agent.context_compressor import (
+        LEGACY_SUMMARY_PREFIX,
+        SUMMARY_PREFIX,
+        _HISTORICAL_SUMMARY_PREFIXES,
+        _content_text_for_contains,
+    )
+
+    text = _content_text_for_contains(content).lstrip()
+    return text.startswith(
+        (SUMMARY_PREFIX, LEGACY_SUMMARY_PREFIX, *_HISTORICAL_SUMMARY_PREFIXES)
+    )
 
 
 def _resolve_to_parent(db, session_id: str) -> tuple[str, bool]:
@@ -259,18 +256,24 @@ def _shape_message(
     is added so callers know the payload was bounded.
     """
     raw_content = m.get("content")
-    if isinstance(raw_content, str) and "\x1b" in raw_content:
+    if max_content_len is not None:
+        from agent.context_compressor import _content_text_for_contains
+
+        shaped_content = _content_text_for_contains(raw_content)
+    else:
+        shaped_content = raw_content
+    if isinstance(shaped_content, str) and "\x1b" in shaped_content:
         # Recalled messages can carry ANSI escape sequences (e.g. archived
         # terminal output). Strip them before returning content to the model.
         from tools.ansi_strip import strip_ansi
 
-        raw_content = strip_ansi(raw_content)
-    if max_content_len and raw_content and len(raw_content) > max_content_len:
-        content = raw_content[:max_content_len] + "…"
+        shaped_content = strip_ansi(shaped_content)
+    if max_content_len and shaped_content and len(shaped_content) > max_content_len:
+        content = shaped_content[:max_content_len] + "…"
         truncated = True
-        original_chars = len(raw_content)
+        original_chars = len(shaped_content)
     else:
-        content = raw_content
+        content = shaped_content
         truncated = False
         original_chars = None
     entry = {
@@ -675,9 +678,20 @@ def _title_match_result(
         "matched_role": "session_title",
         "match_message_id": anchor_id,
         "snippet": f"Session title matched: {session_meta.get('title') or title_query}",
-        "bookend_start": [_shape_message(m) for m in (view.get("bookend_start") or messages[:3])],
-        "messages": [_shape_message(m, anchor_id=anchor_id) for m in (view.get("window") or messages[:5])],
-        "bookend_end": [_shape_message(m) for m in (view.get("bookend_end") or messages[-3:])],
+        "bookend_start": [
+            _shape_message(m, max_content_len=1200)
+            for m in (view.get("bookend_start") or messages[:3])
+            if not _is_compaction_summary(m.get("content", ""))
+        ],
+        "messages": [
+            _shape_message(m, anchor_id=anchor_id, max_content_len=4000)
+            for m in (view.get("window") or messages[:5])
+        ],
+        "bookend_end": [
+            _shape_message(m, max_content_len=1200)
+            for m in (view.get("bookend_end") or messages[-3:])
+            if not _is_compaction_summary(m.get("content", ""))
+        ],
         "messages_before": view.get("messages_before", 0),
         "messages_after": view.get("messages_after", max(len(messages) - 5, 0)),
         "_lineage_root": lineage_root,