fix(title): ignore upload context in conversation titles (#4729)

* fix(title): ignore upload context in conversation titles

* fix(title): cover attachment-only conversations

* fix(title): skip model for attachment-only messages

* fix(title): handle whitespace-only user content

---------

Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
This commit is contained in:
qian 2026-09-02 14:40:12 +08:00 committed by GitHub
parent 3ac40bf9bd
commit 30788c79ff
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 104 additions and 8 deletions

View File

@ -315,7 +315,7 @@ Multi-file upload with automatic document conversion:
- Gateway HTTP uploads stage bytes as `.upload-*.part` files and atomically replace the destination only after size validation. These staging files are hidden from upload listings, agent upload context, and sandbox listing/search tools, and swept on Gateway startup if a hard crash leaves one behind.
- Gateway HTTP upload/list/delete handlers offload filesystem work through `deerflow.utils.file_io.run_file_io`, a dedicated ContextVar-preserving file IO executor. Non-mounted sandbox uploads acquire sandboxes with `SandboxProvider.acquire_async()` and offload `read_bytes()` plus `sandbox.update_file()` together.
- Mounted upload paths skip both sandbox acquisition and per-file synchronization. For AIO remote/provisioner deployments this requires an explicit, accurate `sandbox.thread_data_mounts: true`; omission preserves backend auto-detection.
- Agent receives uploaded file list via `UploadsMiddleware`
- Agent receives uploaded file list via `UploadsMiddleware`; title generation continues to use the original user request rather than the injected upload-context wrapper, with attachment-only messages falling back to `New Conversation`
See [docs/FILE_UPLOAD.md](docs/FILE_UPLOAD.md) for details.

View File

@ -61,7 +61,7 @@ Middlewares execute in strict order, each handling a specific concern:
| 3 | **SandboxMiddleware** | Acquires sandbox environment for code execution |
| 4 | **SummarizationMiddleware** | Reduces context when approaching token limits (optional) |
| 5 | **TodoListMiddleware** | Tracks multi-step tasks in plan mode (optional) |
| 6 | **TitleMiddleware** | Auto-generates conversation titles after first exchange |
| 6 | **TitleMiddleware** | Auto-generates conversation titles from the original user request after first exchange; attachment-only messages fall back to `New Conversation` |
| 7 | **MemoryMiddleware** | Queues conversations for async memory extraction |
| 8 | **ViewImageMiddleware** | Injects image data for vision-capable models (conditional) |
| 9 | **ClarificationMiddleware** | Intercepts clarification requests and interrupts execution (must be last) |

View File

@ -2,6 +2,7 @@
import logging
import re
from collections.abc import Mapping
from typing import TYPE_CHECKING, Any, NotRequired, override
from langchain.agents import AgentState
@ -13,6 +14,7 @@ from langgraph.runtime import Runtime
from deerflow.agents.middlewares.dynamic_context_middleware import is_dynamic_context_reminder
from deerflow.config.title_config import get_title_config
from deerflow.models import create_chat_model
from deerflow.utils.messages import ORIGINAL_USER_CONTENT_KEY, get_original_user_content_text
if TYPE_CHECKING:
from deerflow.config.app_config import AppConfig
@ -106,7 +108,18 @@ class TitleMiddleware(AgentMiddleware[TitleMiddlewareState]):
def _get_title_user_message(self, state: TitleMiddlewareState) -> str:
messages = state.get("messages") or []
user_msg_content = next((self._message_content(m) for m in messages if self._is_user_message_for_title(m)), "")
user_message = next((m for m in messages if self._is_user_message_for_title(m)), None)
if user_message is None:
return ""
if isinstance(user_message, dict):
additional_kwargs = user_message.get("additional_kwargs")
else:
additional_kwargs = getattr(user_message, "additional_kwargs", None)
if isinstance(additional_kwargs, Mapping) and isinstance(additional_kwargs.get(ORIGINAL_USER_CONTENT_KEY), str):
user_msg_content = get_original_user_content_text(self._message_content(user_message), additional_kwargs)
else:
# Keep TitleMiddleware's richer normalization for ordinary structured content.
user_msg_content = self._message_content(user_message)
return self._normalize_content(user_msg_content)
def _should_generate_title(self, state: TitleMiddlewareState, *, allow_partial_exchange: bool = False) -> bool:
@ -170,6 +183,9 @@ class TitleMiddleware(AgentMiddleware[TitleMiddlewareState]):
return title[: config.max_chars] if len(title) > config.max_chars else title
def _fallback_title(self, user_msg: str) -> str:
if not user_msg.strip():
return "New Conversation"
config = self._get_title_config()
fallback_chars = min(config.max_chars, 50)
if len(user_msg) > fallback_chars:
@ -178,7 +194,7 @@ class TitleMiddleware(AgentMiddleware[TitleMiddlewareState]):
ellipsis = "..."
body = min(fallback_chars, config.max_chars - len(ellipsis))
return user_msg[:body].rstrip() + ellipsis
return user_msg if user_msg else "New Conversation"
return user_msg
def _get_runnable_config(self) -> dict[str, Any]:
"""Inherit the parent RunnableConfig and add middleware tag.
@ -217,12 +233,15 @@ class TitleMiddleware(AgentMiddleware[TitleMiddlewareState]):
if not self._should_generate_title(state):
return None
config = self._get_title_config()
if not config.model_name:
user_msg = self._get_title_user_message(state)
user_msg = self._get_title_user_message(state)
# An attachment-only first turn has no user-authored text. Do not let a
# configured title model infer a title from the assistant response.
if not user_msg.strip():
return {"title": self._fallback_title(user_msg)}
user_msg = self._get_title_user_message(state)
config = self._get_title_config()
if not config.model_name:
return {"title": self._fallback_title(user_msg)}
try:
prompt, user_msg = self._build_title_prompt(state)

View File

@ -12,6 +12,7 @@ from deerflow.agents.middlewares import title_middleware as title_middleware_mod
from deerflow.agents.middlewares.dynamic_context_middleware import _DYNAMIC_CONTEXT_REMINDER_KEY
from deerflow.agents.middlewares.title_middleware import TitleMiddleware
from deerflow.config.title_config import TitleConfig, get_title_config, set_title_config
from deerflow.utils.messages import ORIGINAL_USER_CONTENT_KEY
def _clone_title_config(config: TitleConfig) -> TitleConfig:
@ -63,6 +64,82 @@ class TestTitleMiddlewareCoreLogic:
assert middleware._should_generate_title(state) is True
@pytest.mark.parametrize(
"message",
[
HumanMessage(
content="<current_uploads>\nThe following files were uploaded in this message:\n\n- report.pdf\n</current_uploads>\n\n分析这份报告",
additional_kwargs={ORIGINAL_USER_CONTENT_KEY: "分析这份报告"},
),
{
"type": "human",
"content": "<current_uploads>\n...\n</current_uploads>\n\n分析这份报告",
"additional_kwargs": {ORIGINAL_USER_CONTENT_KEY: "分析这份报告"},
},
],
ids=["langchain-message", "checkpoint-dict"],
)
def test_title_uses_original_user_content_when_upload_context_is_injected(self, message):
middleware = TitleMiddleware()
state = {
"messages": [message, AIMessage(content="好的,我来分析报告")],
}
assert middleware._get_title_user_message(state) == "分析这份报告"
def test_attachment_only_title_falls_back_to_new_conversation(self):
_set_test_title_config(enabled=True, model_name=None)
middleware = TitleMiddleware()
state = {
"messages": [
HumanMessage(
content="<current_uploads>\nThe following files were uploaded in this message:\n\n- report.pdf\n</current_uploads>\n\n",
additional_kwargs={ORIGINAL_USER_CONTENT_KEY: ""},
),
AIMessage(content="好的,我来分析 report.pdf"),
]
}
result = asyncio.run(middleware._agenerate_title_result(state))
assert result == {"title": "New Conversation"}
@pytest.mark.parametrize("original_user_content", ["", " \t\n"], ids=["empty", "whitespace-only"])
def test_attachment_only_title_skips_configured_title_model(self, monkeypatch, original_user_content):
_set_test_title_config(enabled=True, model_name="title-model")
middleware = TitleMiddleware()
create_model = MagicMock()
monkeypatch.setattr(title_middleware_module, "create_chat_model", create_model)
state = {
"messages": [
HumanMessage(
content="<current_uploads>\nThe following files were uploaded in this message:\n\n- report.pdf\n</current_uploads>\n\n",
additional_kwargs={ORIGINAL_USER_CONTENT_KEY: original_user_content},
),
AIMessage(content="好的,我来分析 report.pdf"),
]
}
result = asyncio.run(middleware._agenerate_title_result(state))
assert result == {"title": "New Conversation"}
create_model.assert_not_called()
def test_title_preserves_structured_content_without_original_user_content(self):
middleware = TitleMiddleware()
state = {
"messages": [
{
"type": "human",
"content": [{"type": "text", "content": "嵌套的用户请求"}],
},
{"type": "ai", "content": "好的"},
]
}
assert middleware._get_title_user_message(state) == "嵌套的用户请求"
def test_should_not_generate_title_when_disabled_or_already_set(self):
middleware = TitleMiddleware()