fix(5.7): restore subclass cap safety in _truncate_messages path

- Add cap parameter to truncate_messages() so subclass overrides of
  _MAX_SINGLE_CONTENT_CHARS flow through both call paths
- grounding_agent.py delegate forwards self._MAX_SINGLE_CONTENT_CHARS
- Add test_custom_cap_forwarded proving cap propagation
- Found by Opus 4.6 + GPT-5.4 /collab review (both flagged same issue)
- 1,610 passed, 127 skipped

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: Brian Krafft

openspace/agents/grounding/messages.py

@@ -58,6 +58,7 @@ def truncate_messages(
     messages: List[Dict[str, Any]],
     keep_recent: int = 8,
     max_tokens_estimate: int = 120_000,
+    cap: int = _MAX_SINGLE_CONTENT_CHARS,
 ) -> List[Dict[str, Any]]:
     """Trim conversation history to fit within token budget.
 
@@ -71,11 +72,12 @@ def truncate_messages(
         messages: Full message list.
         keep_recent: Number of recent conversation rounds to preserve.
         max_tokens_estimate: Approximate token budget.
+        cap: Per-message character cap (forwarded to cap_message_content).
 
     Returns:
         Possibly shortened message list.
     """
-    messages = cap_message_content(messages)
+    messages = cap_message_content(messages, cap)
 
     if len(messages) <= keep_recent + 2:  # +2 for system and initial user
         return messages

openspace/agents/grounding_agent.py

@@ -122,7 +122,9 @@ def _cap_message_content(cls, messages: List[Dict[str, Any]]) -> List[Dict[str,
     def _truncate_messages(
         self, messages: List[Dict[str, Any]], keep_recent: int = 8, max_tokens_estimate: int = 120000
     ) -> List[Dict[str, Any]]:
-        return _truncate_messages_impl(messages, keep_recent, max_tokens_estimate)
+        return _truncate_messages_impl(
+            messages, keep_recent, max_tokens_estimate, cap=self._MAX_SINGLE_CONTENT_CHARS
+        )
 
     async def process(self, context: Dict[str, Any]) -> Dict[str, Any]:
         """

tests/test_grounding_messages.py

@@ -139,6 +139,21 @@ def test_default_keep_recent_is_8(self):
         # system(1) + user(1) + recent(16) = 18
         assert len(result) == 18
 
+    def test_custom_cap_forwarded(self):
+        """Verify the cap parameter flows through to cap_message_content."""
+        msgs = [
+            {"role": "system", "content": "sys"},
+            {"role": "user", "content": "go"},
+        ]
+        # Add enough oversized messages to trigger truncation
+        for i in range(20):
+            msgs.append({"role": "assistant", "content": "x" * 500})
+        result = truncate_messages(msgs, keep_recent=2, max_tokens_estimate=1, cap=200)
+        # Messages should have been capped at 200 chars
+        for m in result:
+            if m["role"] != "system" and m["role"] != "user":
+                assert len(m["content"]) <= 250  # some overhead from truncation marker
+
 
 # ── Delegation seam tests ──────────────────────────────────────────