From 9220a876bcc8f2a651212ed8a3955ecb602fdcad Mon Sep 17 00:00:00 2001
From: Hermes Agent <agent@hermes>
Date: Wed, 15 Apr 2026 16:57:31 +0000
Subject: [PATCH 1/2] fix: strip orphaned tool messages before sending history
 to API (fixes #534)

Extends _sanitize_messages_for_api() with a two-pass approach:
1. Collect all tool_call_ids declared in assistant messages (handles
   both OpenAI 'id' and Anthropic 'call_id' field names).
2. Drop any tool-role messages whose tool_call_id was not declared
   by a preceding assistant message.

Strictly-conformant providers (Mercury-2/Inception, newer OpenAI
models) reject histories with orphaned tool results with a 400 error:
'Message has tool role, but there was no previous assistant message
with a tool call.' This can happen when histories are edited, when
switching between providers, or when partial messages are stored.

Adds 13 regression tests covering: valid roundtrip preservation,
multiple tool calls, partial orphan filtering, Anthropic call_id,
edge cases (None tool_calls, missing tool_call_id, non-dict entries).
---
 api/streaming.py                     |  26 ++++
 tests/test_orphaned_tool_messages.py | 175 +++++++++++++++++++++++++++
 2 files changed, 201 insertions(+)
 create mode 100644 tests/test_orphaned_tool_messages.py

diff --git a/api/streaming.py b/api/streaming.py
index 23189a8..f77ce82 100644
--- a/api/streaming.py
+++ b/api/streaming.py
@@ -64,11 +64,37 @@ def _sanitize_messages_for_api(messages):
     The webui stores extra metadata on messages (attachments, timestamp, _ts)
     for display purposes. Some providers (e.g. Z.AI/GLM) reject unknown fields
     instead of ignoring them, causing HTTP 400 errors on subsequent messages.
+
+    Also strips orphaned tool-role messages whose tool_call_id cannot be linked
+    to a preceding assistant message with tool_calls. Strictly-conformant providers
+    (Mercury-2/Inception, newer OpenAI models) reject histories containing dangling
+    tool results with a 400 error: "Message has tool role, but there was no previous
+    assistant message with a tool call."
     """
+    # First pass: collect all tool_call_ids declared by assistant messages.
+    # Handles both OpenAI ('id') and Anthropic ('call_id') field names.
+    valid_tool_call_ids: set = set()
+    for msg in messages:
+        if not isinstance(msg, dict):
+            continue
+        if msg.get('role') == 'assistant':
+            for tc in msg.get('tool_calls') or []:
+                if isinstance(tc, dict):
+                    tid = tc.get('id') or tc.get('call_id') or ''
+                    if tid:
+                        valid_tool_call_ids.add(tid)
+
+    # Second pass: build the sanitized list, dropping orphaned tool messages.
     clean = []
     for msg in messages:
         if not isinstance(msg, dict):
             continue
+        role = msg.get('role')
+        if role == 'tool':
+            tid = msg.get('tool_call_id') or ''
+            if not tid or tid not in valid_tool_call_ids:
+                # Orphaned tool result — skip to avoid 400 from strict providers.
+                continue
         sanitized = {k: v for k, v in msg.items() if k in _API_SAFE_MSG_KEYS}
         if sanitized.get('role'):
             clean.append(sanitized)
diff --git a/tests/test_orphaned_tool_messages.py b/tests/test_orphaned_tool_messages.py
new file mode 100644
index 0000000..7df71ec
--- /dev/null
+++ b/tests/test_orphaned_tool_messages.py
@@ -0,0 +1,175 @@
+"""Tests for _sanitize_messages_for_api() orphaned-tool-message stripping.
+
+Regression for issue #534: strictly-conformant providers (Mercury-2/Inception,
+newer OpenAI models) reject histories containing tool-role messages whose
+tool_call_id has no matching tool_calls entry in a prior assistant message.
+"""
+import sys
+import pathlib
+
+REPO_ROOT = pathlib.Path(__file__).parent.parent.resolve()
+sys.path.insert(0, str(REPO_ROOT))
+
+from api.streaming import _sanitize_messages_for_api
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _asst_with_tool_call(call_id="call-1", call_id_key="id"):
+    return {
+        "role": "assistant",
+        "content": None,
+        "tool_calls": [{"type": "function", call_id_key: call_id, "function": {"name": "terminal", "arguments": "{}"}}],
+        "_ts": 12345,  # extra field that should be stripped
+    }
+
+
+def _tool_result(call_id="call-1"):
+    return {"role": "tool", "tool_call_id": call_id, "content": "ok", "_ts": 12345}
+
+
+def _user(text="hello"):
+    return {"role": "user", "content": text, "_ts": 12345}
+
+
+def _asst(text="hi"):
+    return {"role": "assistant", "content": text, "_ts": 12345}
+
+
+# ---------------------------------------------------------------------------
+# Tests: normal valid histories are preserved
+# ---------------------------------------------------------------------------
+
+def test_valid_tool_roundtrip_preserved():
+    """A linked assistant→tool pair must be kept intact."""
+    msgs = [_user(), _asst_with_tool_call("call-1"), _tool_result("call-1"), _asst()]
+    result = _sanitize_messages_for_api(msgs)
+    roles = [m["role"] for m in result]
+    assert roles == ["user", "assistant", "tool", "assistant"]
+
+
+def test_extra_fields_stripped():
+    """Non-API fields (_ts etc.) are always stripped."""
+    msgs = [_user(), _asst()]
+    result = _sanitize_messages_for_api(msgs)
+    for m in result:
+        assert "_ts" not in m
+
+
+def test_valid_history_without_tool_messages_unchanged():
+    """Plain user/assistant history with no tool calls is passed through unchanged."""
+    msgs = [_user("a"), _asst("b"), _user("c"), _asst("d")]
+    result = _sanitize_messages_for_api(msgs)
+    assert len(result) == 4
+    assert all(m["role"] in ("user", "assistant") for m in result)
+
+
+def test_multiple_valid_tool_calls_preserved():
+    """Multiple linked tool_call_ids in one assistant message are all preserved."""
+    asst = {
+        "role": "assistant",
+        "content": None,
+        "tool_calls": [
+            {"type": "function", "id": "call-1", "function": {"name": "f1", "arguments": "{}"}},
+            {"type": "function", "id": "call-2", "function": {"name": "f2", "arguments": "{}"}},
+        ],
+    }
+    msgs = [_user(), asst, _tool_result("call-1"), _tool_result("call-2"), _asst()]
+    result = _sanitize_messages_for_api(msgs)
+    roles = [m["role"] for m in result]
+    assert roles == ["user", "assistant", "tool", "tool", "assistant"]
+
+
+# ---------------------------------------------------------------------------
+# Tests: orphaned tool messages are dropped
+# ---------------------------------------------------------------------------
+
+def test_orphaned_tool_message_dropped():
+    """A tool message with no matching assistant tool_call is dropped."""
+    msgs = [_user(), _asst(), _tool_result("call-orphan")]
+    result = _sanitize_messages_for_api(msgs)
+    roles = [m["role"] for m in result]
+    assert "tool" not in roles
+    assert roles == ["user", "assistant"]
+
+
+def test_tool_message_missing_tool_call_id_dropped():
+    """A tool message with no tool_call_id at all is dropped."""
+    msg = {"role": "tool", "content": "result"}
+    msgs = [_user(), _asst_with_tool_call("call-1"), msg]
+    result = _sanitize_messages_for_api(msgs)
+    roles = [m["role"] for m in result]
+    assert "tool" not in roles
+
+
+def test_partially_orphaned_tool_messages():
+    """In a mixed batch, only the orphaned tool messages are dropped."""
+    asst = _asst_with_tool_call("call-valid")
+    msgs = [
+        _user(),
+        asst,
+        _tool_result("call-valid"),   # linked → kept
+        _tool_result("call-ghost"),   # orphaned → dropped
+        _asst(),
+    ]
+    result = _sanitize_messages_for_api(msgs)
+    roles = [m["role"] for m in result]
+    assert roles == ["user", "assistant", "tool", "assistant"]
+    # The kept tool message has the right call_id
+    tool_msgs = [m for m in result if m["role"] == "tool"]
+    assert tool_msgs[0]["tool_call_id"] == "call-valid"
+
+
+def test_orphaned_tool_only_history():
+    """A history consisting only of orphaned tool messages returns empty."""
+    msgs = [_tool_result("dangling-1"), _tool_result("dangling-2")]
+    result = _sanitize_messages_for_api(msgs)
+    assert result == []
+
+
+# ---------------------------------------------------------------------------
+# Tests: Anthropic 'call_id' field name (not OpenAI 'id')
+# ---------------------------------------------------------------------------
+
+def test_anthropic_call_id_field_recognized():
+    """Anthropic tool calls use 'call_id' not 'id' — both must be recognized."""
+    asst = _asst_with_tool_call("call-anthropic", call_id_key="call_id")
+    msgs = [_user(), asst, _tool_result("call-anthropic"), _asst()]
+    result = _sanitize_messages_for_api(msgs)
+    roles = [m["role"] for m in result]
+    assert roles == ["user", "assistant", "tool", "assistant"]
+
+
+# ---------------------------------------------------------------------------
+# Tests: edge cases
+# ---------------------------------------------------------------------------
+
+def test_empty_messages_list():
+    assert _sanitize_messages_for_api([]) == []
+
+
+def test_non_dict_messages_skipped():
+    """Non-dict items in the messages list are silently ignored."""
+    msgs = ["not a dict", None, _user("hi"), 42]
+    result = _sanitize_messages_for_api(msgs)
+    assert len(result) == 1
+    assert result[0]["role"] == "user"
+
+
+def test_tool_calls_none_does_not_crash():
+    """An assistant message with tool_calls=None is handled without crashing."""
+    asst = {"role": "assistant", "content": "hello", "tool_calls": None}
+    msgs = [_user(), asst, _tool_result("call-1")]
+    result = _sanitize_messages_for_api(msgs)
+    # call-1 has no valid parent (tool_calls=None → no IDs registered) → dropped
+    roles = [m["role"] for m in result]
+    assert "tool" not in roles
+
+
+def test_system_messages_preserved():
+    """System messages are always preserved."""
+    msgs = [{"role": "system", "content": "You are helpful."}, _user(), _asst()]
+    result = _sanitize_messages_for_api(msgs)
+    assert result[0]["role"] == "system"

From ce4e01ea9272fa54ca69447731ccf9b4c0afcb28 Mon Sep 17 00:00:00 2001
From: Hermes Agent <agent@hermes>
Date: Wed, 15 Apr 2026 19:06:54 +0000
Subject: [PATCH 2/2] chore: bump version to v0.50.51, update CHANGELOG

---
 CHANGELOG.md      | 5 +++++
 static/index.html | 2 +-
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index f33c9a2..6602d46 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,10 @@
 # Hermes Web UI -- Changelog
 
+## [v0.50.51] — 2026-04-15
+
+### Fixed
+- **Orphaned tool messages** — conversation histories containing `role: tool` messages with no matching `tool_call_id` in a prior assistant message are now silently stripped before sending to the provider API. Fixes 400 errors from strictly-conformant providers (Mercury-2/Inception, newer OpenAI models). Adds 13 regression tests. (Fixes #534)
+
 ## [v0.50.50] — 2026-04-15
 
 ### Fixed
diff --git a/static/index.html b/static/index.html
index ef38643..347b366 100644
--- a/static/index.html
+++ b/static/index.html
@@ -552,7 +552,7 @@
                 <div class="settings-section-title">System</div>
                 <div class="settings-section-meta">Instance version and access controls.</div>
               </div>
-              <span class="settings-version-badge">v0.50.50</span>
+              <span class="settings-version-badge">v0.50.51</span>
             </div>
             <div class="settings-field" style="border-top:1px solid var(--border);padding-top:12px;margin-top:8px">
               <label for="settingsPassword" data-i18n="settings_label_password">Access Password</label>