Skip to content

ccran / lufa-contract

Go to a project
Commit 68c08496 by ccran
Options

feat: 添加反思；添加多合同支持；

parent 9fa3fd96
Showing with 90 additions and 119 deletions
core/memory.py
......@@ -262,8 +262,8 @@ class MemoryStore:
with self._lock:
wb = Workbook()
ws_findings = wb.active
ws_findings.title = "findings"
ws_final_findings = wb.active
ws_final_findings.title = "findings"
finding_headers = [
("rule_title", "规则标题"),
......@@ -273,19 +273,19 @@ class MemoryStore:
("risk_level", "风险等级"),
("suggestion", "建议"),
]
ws_findings.append([label for _, label in finding_headers])
for f in self.findings:
ws_findings.append([
getattr(f, key, "") for key, _ in finding_headers
])
ws_final_findings = wb.create_sheet("final_findings")
ws_final_findings.append([label for _, label in finding_headers])
for f in self.final_findings:
ws_final_findings.append([
getattr(f, key, "") for key, _ in finding_headers
])
ws_findings = wb.create_sheet("findings")
ws_findings.append([label for _, label in finding_headers])
for f in self.findings:
ws_findings.append([
getattr(f, key, "") for key, _ in finding_headers
])
ws_facts = wb.create_sheet("facts")
if self.facts:
fact_keys: List[str] = sorted({k for item in self.facts for k in item.keys()})
......@@ -321,6 +321,7 @@ class MemoryStore:
doc_obj: DocBase,
file_name: Optional[str] = None,
remove_prefix: bool = False,
export_final: bool = False,
) -> Dict[str, Any]:
"""Add all findings as comments to a document, upload, then delete the local file."""
if doc_obj is None:
......@@ -334,11 +335,14 @@ class MemoryStore:
name = f"{name}{suffix}"
output_path = Path(__file__).resolve().parent.parent / "tmp" / name
if export_final:
target_findings = self.final_findings
else:
target_findings = self.findings
with self._lock:
comments: List[Dict[str, Any]] = []
for idx, f in enumerate(self.findings, start=1):
# 导出final_findings
# for idx, f in enumerate(self.final_findings, start=1):
for idx, f in enumerate(target_findings, start=1):
segment_id = int(f.segment_id or 0)
chunk_id = max(segment_id - 1, 0)
suggest_parts = []
......
core/tools/reflect_retry.py
......@@ -8,21 +8,12 @@ from core.tools.segment_llm import LLMTool
REFLECT_SYSTEM_PROMPT = '''
你是一个合同审查反思智能体（ReviewReflection）。
你要基于 facts 与全文上下文，对已有 findings 进行校正后，输出【最终可交付的 findings 列表】。
你的任务是：基于 facts 与全文上下文，
对已有 findings 输出修改操作。
你只能对 findings 执行以下四种操作：
- keep：确认该风险结论成立且无需修改
- update：修改一个已有风险
- add：新增一个由全文结构推导出的风险
- remove：删除一个不成立的风险
【输出约束】
- 输出必须是 JSON
- 每条操作必须仅包含字段：op、id、findings
- findings 在 add / update 时必须是完整 finding
要求：
- 严格按照输出 JSON Schema 返回结果，不得输出任何解释性文字
- 最终 findings 中每条都必须证据充分，original_text 必须是合同原文直接引用
- 不得引入新的审查维度，只能基于已有 findings 的范围做合并、修订、删除或系统性总结
'''
REFLECT_USER_PROMPT = '''
......@@ -36,30 +27,25 @@ REFLECT_USER_PROMPT = '''
【合同立场】
站在 {party_role} 的立场进行反思审查。
【反思原则】
- 不得引入新的审查维度
- 所有判断必须有 facts 或合同原文证据支持
- 仅对已有 findings 进行增add、改update、删remove、保留keep操作
- 若风险在全文上下文中不成立，必须 remove
- 若风险成立但表述、严重性或建议不准确，必须 update
- 若由多个 findings 或全文结构推导出新的系统性风险，可 add
【输出要求】
- 输出必须是 JSON
- 每条操作必须仅包含字段：op、id、findings
- findings 在 add / update 时必须是完整 finding
【任务】
输出反思后的最终 findings 列表（可直接用于最终审查报告）：
- 删除在全文上下文中不成立的 findings
- 修订表述/严重性/建议不准确的 findings
- 如需合并重复 findings，请合并成一条（保留全部原文证据引用）
- 如可由全文结构推导出系统性风险，可新增 1~3 条 global findings（仍需原文证据）
- 严格按照输出 JSON Schema 返回结果，不得输出任何解释性文字
'''
OUTPUT_FORMAT_SCHEMA = '''
```json
{
"operations": [
{
"op": "keep | add | update | remove",
"id": "string | null",
"findings": "object | null"
}
]
"final_findings": [
{
"segment_id":"合同原文片段所在的段落ID",
"issue": "详细的风险描述",
"original_text": "合同原文片段的直接引用",
"suggestion": "可直接替换原文或新增的条款措辞"
}
]
}
```
'''
......@@ -98,9 +84,13 @@ class ReflectRetryTool(LLMTool):
except Exception:
data = {}
operations = data.get("operations", []) or []
final_findings = self._apply_operations(base_findings, operations)
final_findings = data.get("final_findings", []) or []
for finding in final_findings:
try:
finding['segment_id'] = int(finding.get('segment_id', 0))
except Exception:
finding['segment_id'] = 0
finding['rule_title'] = rule.get('title','')
return final_findings
def _build_findings_with_ids(self, findings: List[Dict]) -> List[Dict[str, Any]]:
......@@ -112,33 +102,6 @@ class ReflectRetryTool(LLMTool):
res.append(item)
return res
def _apply_operations(self, base_findings: List[Dict[str, Any]], operations: List[Dict]) -> List[Dict[str, Any]]:
by_id = {str(f.get("id")): dict(f) for f in base_findings if f.get("id") is not None}
added: List[Dict[str, Any]] = []
for op in operations:
action = (op.get("op") or "").strip().lower()
target_id = op.get("id")
payload = op.get("findings")
if action == "keep":
if target_id is not None and str(target_id) in by_id:
continue
elif action == "remove":
if target_id is not None:
by_id.pop(str(target_id), None)
elif action == "update":
if target_id is not None and isinstance(payload, dict):
payload = dict(payload)
payload["id"] = str(target_id)
by_id[str(target_id)] = payload
elif action == "add":
if isinstance(payload, dict):
added.append(dict(payload))
merged = list(by_id.values()) + added
return merged
if __name__ == "__main__":
tool = ReflectRetryTool()
......
core/tools/segment_review.py
......@@ -10,6 +10,9 @@ from utils.excel_util import ExcelUtil
from core.tools.segment_llm import LLMTool
import re
DEFAULT_RULESET_ID = "通用"
ALL_RULESET_IDS = ["通用","借款","担保","测试"]
REVIEW_SYSTEM_PROMPT = '''
你是一个专业的合同分段审查智能体（SegmentReview）。
你的核心任务是对“当前分段”进行【法律风险识别】，并给出可落地的修改建议。
......@@ -103,7 +106,7 @@ def _is_generic_suggestion(text: str) -> bool:
class SegmentReviewTool(LLMTool):
def __init__(self):
super().__init__(REVIEW_SYSTEM_PROMPT)
self.rule_version = "通用"
self.default_ruleset_id = DEFAULT_RULESET_ID
self.column_map = {
"id": "ID",
"title": "审查项",
......@@ -113,8 +116,10 @@ class SegmentReviewTool(LLMTool):
"suggestion_template": "建议模板",
}
rules_path = Path(__file__).resolve().parent.parent.parent / "data" / "rules.xlsx"
rules = ExcelUtil.load_mapped_excel(rules_path, sheet_name=self.rule_version, column_map=self.column_map)
self.rulesets: Dict[str, List[Dict]] = {self.rule_version: rules} if rules else {}
self.rulesets: Dict[str, List[Dict]] = {}
for rs_id in ALL_RULESET_IDS:
rules = ExcelUtil.load_mapped_excel(rules_path, sheet_name=rs_id, column_map=self.column_map)
self.rulesets[rs_id] = rules
@tool_func(
{
......@@ -130,7 +135,7 @@ class SegmentReviewTool(LLMTool):
}
)
def run(self, segment_id: str, segment_text: str, ruleset_id: str, party_role: str, context_memories: Optional[List[Dict]] = None) -> Dict:
rules = self.rulesets.get(ruleset_id) or self.rulesets.get(self.rule_version, []) or []
rules = self.rulesets.get(ruleset_id) or self.rulesets.get(self.default_ruleset_id, []) or []
result = self._evaluate_rules(party_role,segment_id,segment_text,rules, context_memories)
overall = "revise" if (result["findings"] ) else "pass"
......
data/rules.xlsx
No preview for this file type
main.py
......@@ -67,7 +67,7 @@ async def parse_document(payload: DocumentParseRequest) -> DocumentParseResponse
# text = doc_obj.get_all_text()
chunk_ids = doc_obj.get_chunk_id_list()
# get ruleset items
ruleset_id = payload.ruleset_id or review_tool.rule_version
ruleset_id = payload.ruleset_id or review_tool.default_ruleset_id
ruleset_items = review_tool.rulesets.get(ruleset_id) or []
ruleset_review_items = [r.get('title') for r in ruleset_items]
return DocumentParseResponse(
......@@ -197,41 +197,39 @@ class ReflectReviewResponse(BaseModel):
@app.post("/segments/review/reflect", response_model=ReflectReviewResponse)
def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse:
store = get_cached_memory(payload.conversation_id)
ruleset_id = payload.ruleset_id or review_tool.rule_version
ruleset_items = review_tool.rulesets.get(ruleset_id) or []
rule = next((r for r in ruleset_items if r.get("title") == payload.rule_title), None)
if not rule:
raise HTTPException(status_code=404, detail=f"Rule not found: {payload.rule_title}")
facts = store.get_facts()
findings = [f.__dict__ for f in store.list_findings()]
final_findings = reflect_tool.run(
party_role=payload.party_role,
rule=rule,
facts=facts,
findings=findings,
)
for f in final_findings or []:
try:
store.add_final_finding_from_dict({
"rule_title": f.get("rule_title", ""),
"segment_id": f.get("segment_id", 0),
"original_text": f.get("original_text", ""),
"issue": f.get("issue", ""),
"risk_level": (f.get("risk_level") or f.get("level") or "").upper(),
"suggestion": f.get("suggestion", ""),
})
except Exception:
continue
return ReflectReviewResponse(
conversation_id=payload.conversation_id,
rule_title=payload.rule_title,
findings=final_findings or [],
)
store = get_cached_memory(payload.conversation_id)
ruleset_id = payload.ruleset_id or review_tool.default_ruleset_id
ruleset_items = review_tool.rulesets.get(ruleset_id) or []
rule = next((r for r in ruleset_items if r.get("title") == payload.rule_title), None)
if not rule:
raise HTTPException(status_code=404, detail=f"Rule not found: {payload.rule_title}")
facts = store.get_facts()
findings = [f.__dict__ for f in store.search_findings("", rule_title=payload.rule_title)]
final_findings = reflect_tool.run(
party_role=payload.party_role,
rule=rule,
facts=facts,
findings=findings,
)
for f in final_findings or []:
try:
store.add_final_finding_from_dict({
"rule_title": f.get("rule_title", ""),
"segment_id": f.get("segment_id", 0),
"original_text": f.get("original_text", ""),
"issue": f.get("issue", ""),
"risk_level": (f.get("risk_level") or f.get("level") or "H").upper(),
"suggestion": f.get("suggestion", ""),
})
# print(f'len(store) final_findings:{len(store.final_findings)}')
except Exception:
continue
return ReflectReviewResponse(
conversation_id=payload.conversation_id,
rule_title=payload.rule_title,
findings=final_findings or [],
)
########################################################################################################################
class ConversationResponse(BaseModel):
......@@ -250,6 +248,7 @@ class MemoryExportRequest(BaseModel):
conversation_id: str
file_ext: str
file_name: Optional[str] = None
export_final: Optional[bool] = False
class MemoryExportResponse(BaseModel):
......@@ -273,7 +272,7 @@ def export_memory(payload: MemoryExportRequest) -> MemoryExportResponse:
raise HTTPException(status_code=500, detail=f"Export failed: {exc}")
try:
doc_res = store.export_findings_to_doc_comments(doc_obj)
doc_res = store.export_findings_to_doc_comments(doc_obj,export_final=payload.export_final or False)
except Exception as exc:
traceback.print_exc()
raise HTTPException(status_code=500, detail=f"Export doc comments failed: {exc}")
......@@ -285,6 +284,6 @@ def export_memory(payload: MemoryExportRequest) -> MemoryExportResponse:
)
if __name__ == "__main__":
uvicorn.run(
"main:app", host="0.0.0.0", port=18168, log_level="info", reload=False
)
\ No newline at end of file
uvicorn.run(
"main:app", host="0.0.0.0", port=18168, log_level="info", reload=False
)
\ No newline at end of file
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or sign in to comment