提交代码

This commit is contained in:
董宏凯 committed 2026-07-01 19:24:23 +08:00
1 parent d21dbeb214
commit 54190d7a83
8 files changed
+1205 -11

No files matched your search

@@ -25,15 +25,18 @@ import re
import time
from types import SimpleNamespace
from langchain_core.messages import HumanMessage, SystemMessage
from langchain_core.runnables import RunnableConfig
from ..search import get_search_provider, SearchResult
from ..prompts.researcher_prompts import (
EXTRACT_KEYWORDS_SYSTEM,
EXTRACT_KEYWORDS_USER,
RANK_MATERIALS_SYSTEM,
RANK_MATERIALS_USER,
)
from ..search import SearchResult, get_search_provider
from ..search.skill_search import SkillPicker
from ..state import AIWritingState, Material, MaterialPackage
from ..prompts.researcher_prompts import (
EXTRACT_KEYWORDS_SYSTEM, EXTRACT_KEYWORDS_USER,
RANK_MATERIALS_SYSTEM, RANK_MATERIALS_USER,
)
from ._agent_runner import (
KEYWORDS_READY_MARKER,
MATERIALS_RANKED_MARKER,
@@ -45,7 +48,16 @@ from ._agent_runner import (
use_builtin_agent,
)
from ._skill_agent import run_skill_via_real_agent, use_skill_search_agent
from ._utils import get_ai_writing_config, llm_json, safe_get_stream_writer
from ._utils import (
ThinkTagParser,
_content_to_text,
get_ai_writing_config,
get_model,
llm_json,
parse_json,
reasoning_delta,
safe_get_stream_writer,
)
logger = logging.getLogger(__name__)
@@ -227,6 +239,40 @@ def _per_keyword_limit(max_materials: int, total_keywords: int) -> int:
return (max_materials + total_keywords - 1) // total_keywords
async def _stream_llm_json(
system: str,
user: str,
*,
fallback: dict,
config: RunnableConfig,
model_name: str | None,
disable_thinking: bool,
on_text,
) -> dict:
"""流式生成一段 JSON,并把正文 / 思考片段透传给前端过程卡。"""
messages = [SystemMessage(content=system), HumanMessage(content=user)]
full_text = ""
parser = ThinkTagParser()
try:
model = get_model(config, model_name=model_name, disable_thinking=disable_thinking)
async for chunk in model.astream(messages):
reasoning = reasoning_delta(chunk)
if reasoning:
on_text(reasoning, True)
delta = _content_to_text(chunk.content)
if not delta:
continue
for is_thinking, text in parser.feed(delta):
if not is_thinking:
full_text += text
if text:
on_text(text, is_thinking)
return parse_json(full_text, fallback)
except Exception as e:
logger.error("stream_llm_json failed: %s", e, exc_info=True)
return {**fallback, "_llm_error": str(e)}
# 「再追加 N 条素材」之类的数量指令:动词 + 数字(+ 量词 + 名词)。
_APPEND_COUNT_VERBS = "追加|新增|增加|补充|再来|多来|多给|多找|多搜|多检索|再给|再要|再找|再搜|再加"
_APPEND_COUNT_RE = re.compile(
@@ -467,6 +513,15 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
def _emit(message: str) -> None:
writer({"type": "research_progress", "message": message})
def _emit_keyword_chunk(text: str = "", is_thinking: bool = False) -> None:
writer({
"type": "keyword_chunk",
"agent_name": researcher_label,
"message": "生成检索词",
"text": text,
"is_thinking": is_thinking,
})
# 跳过素材收集:用户在初始表单勾选「不收集素材,直接用通用知识写作」。
# 仅在**首次**进入(无任何 pause 干预 = intervention 为空)时短路,避免后续
# 「补充检索 / 按输入追加」等再入研究节点时被误跳过。直接产出空素材包、强制
@@ -501,6 +556,7 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
_emit(f"追加检索 {len(keywords)} 个关键词…")
else:
_emit("正在生成检索词…")
_emit_keyword_chunk()
if use_agent:
kw_result = await run_writing_agent(
agent_id=RESEARCHER_AGENT_ID,
@@ -514,12 +570,13 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
fallback={"keywords": [intent]},
state=state,
config=config,
on_text=_emit_keyword_chunk,
explicit_model_fallback=keyword_model,
strict_model=False,
disable_thinking=True,
)
else:
kw_result = await llm_json(
kw_result = await _stream_llm_json(
EXTRACT_KEYWORDS_SYSTEM,
EXTRACT_KEYWORDS_USER.format(
user_intent=intent,
@@ -531,6 +588,7 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
config=config,
model_name=keyword_model,
disable_thinking=True,
on_text=_emit_keyword_chunk,
)
# 清洗关键词:剔除占位符 / 空值,全空时回退写作意图
keywords = _clean_keywords(kw_result.get("keywords"), intent)