feat: enhance agent orchestrator with runtime hardening and tool governance [AC-MARH-01~12] [AC-IDMP-11~18]
This commit is contained in:
MerCry
parent
978aaee885
commit
5f4bde8752
|
|
@ -933,6 +933,7 @@ async def _execute_agent_mode(
|
|||
timeout_governor=timeout_governor,
|
||||
llm_client=llm_client,
|
||||
tool_registry=tool_registry,
|
||||
tenant_id=tenant_id,
|
||||
)
|
||||
|
||||
final_answer, react_ctx, agent_trace = await orchestrator.execute(
|
||||
|
|
|
|||
|
|
@ -5,6 +5,7 @@ Middleware for AI Service.
|
|||
|
||||
import logging
|
||||
import re
|
||||
import uuid
|
||||
from collections.abc import Callable
|
||||
|
||||
from fastapi import Request, Response, status
|
||||
|
|
@ -20,6 +21,17 @@ TENANT_ID_HEADER = "X-Tenant-Id"
|
|||
API_KEY_HEADER = "X-API-Key"
|
||||
ACCEPT_HEADER = "Accept"
|
||||
SSE_CONTENT_TYPE = "text/event-stream"
|
||||
REQUEST_ID_HEADER = "X-Request-Id"
|
||||
|
||||
# Prompt template protected variable names injected by system/runtime.
|
||||
# These are reserved for internal orchestration and should not be overridden by user input.
|
||||
PROMPT_PROTECTED_VARIABLES = {
|
||||
"available_tools",
|
||||
"query",
|
||||
"history",
|
||||
"internal_protocol",
|
||||
"output_contract",
|
||||
}
|
||||
|
||||
TENANT_ID_PATTERN = re.compile(r'^[^@]+@ash@\d{4}$')
|
||||
|
||||
|
|
@ -29,6 +41,15 @@ PATHS_SKIP_API_KEY = {
|
|||
"/docs",
|
||||
"/redoc",
|
||||
"/openapi.json",
|
||||
"/favicon.ico",
|
||||
"/openapi/v1/share/chat",
|
||||
}
|
||||
|
||||
PATHS_SKIP_TENANT = {
|
||||
"/health",
|
||||
"/ai/health",
|
||||
"/favicon.ico",
|
||||
"/openapi/v1/share/chat",
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -63,17 +84,24 @@ class ApiKeyMiddleware(BaseHTTPMiddleware):
|
|||
if self._should_skip_api_key(request.url.path):
|
||||
return await call_next(request)
|
||||
|
||||
request_id = request.headers.get(REQUEST_ID_HEADER) or str(uuid.uuid4())
|
||||
request.state.request_id = request_id
|
||||
|
||||
api_key = request.headers.get(API_KEY_HEADER)
|
||||
|
||||
if not api_key or not api_key.strip():
|
||||
logger.warning(f"[AC-AISVC-50] Missing X-API-Key header for {request.url.path}")
|
||||
return JSONResponse(
|
||||
logger.warning(
|
||||
f"[AC-AISVC-50] Missing X-API-Key header for {request.url.path}, request_id={request_id}"
|
||||
)
|
||||
response = JSONResponse(
|
||||
status_code=status.HTTP_401_UNAUTHORIZED,
|
||||
content=ErrorResponse(
|
||||
code=ErrorCode.UNAUTHORIZED.value,
|
||||
message="Missing required header: X-API-Key",
|
||||
).model_dump(exclude_none=True),
|
||||
)
|
||||
response.headers[REQUEST_ID_HEADER] = request_id
|
||||
return response
|
||||
|
||||
api_key = api_key.strip()
|
||||
|
||||
|
|
@ -90,17 +118,45 @@ class ApiKeyMiddleware(BaseHTTPMiddleware):
|
|||
except Exception as e:
|
||||
logger.error(f"[AC-AISVC-50] Failed to initialize API key service: {e}")
|
||||
|
||||
if not service.validate_key(api_key):
|
||||
logger.warning(f"[AC-AISVC-50] Invalid API key for {request.url.path}")
|
||||
return JSONResponse(
|
||||
client_ip = request.client.host if request.client else None
|
||||
tenant_id = request.headers.get(TENANT_ID_HEADER, "")
|
||||
|
||||
validation = service.validate_key_with_context(api_key, client_ip=client_ip)
|
||||
if not validation.ok:
|
||||
if validation.reason == "rate_limited":
|
||||
logger.warning(
|
||||
f"[AC-AISVC-50] Rate limited: path={request.url.path}, tenant={tenant_id}, "
|
||||
f"ip={client_ip}, qpm={validation.rate_limit_qpm}, request_id={request_id}"
|
||||
)
|
||||
response = JSONResponse(
|
||||
status_code=status.HTTP_429_TOO_MANY_REQUESTS,
|
||||
content=ErrorResponse(
|
||||
code=ErrorCode.SERVICE_UNAVAILABLE.value,
|
||||
message="Rate limit exceeded",
|
||||
details=[{"reason": "rate_limited", "limit_qpm": validation.rate_limit_qpm}],
|
||||
).model_dump(exclude_none=True),
|
||||
)
|
||||
response.headers[REQUEST_ID_HEADER] = request_id
|
||||
return response
|
||||
|
||||
logger.warning(
|
||||
f"[AC-AISVC-50] API key validation failed: reason={validation.reason}, "
|
||||
f"path={request.url.path}, tenant={tenant_id}, ip={client_ip}, request_id={request_id}"
|
||||
)
|
||||
response = JSONResponse(
|
||||
status_code=status.HTTP_401_UNAUTHORIZED,
|
||||
content=ErrorResponse(
|
||||
code=ErrorCode.UNAUTHORIZED.value,
|
||||
message="Invalid API key",
|
||||
details=[{"reason": validation.reason}],
|
||||
).model_dump(exclude_none=True),
|
||||
)
|
||||
response.headers[REQUEST_ID_HEADER] = request_id
|
||||
return response
|
||||
|
||||
return await call_next(request)
|
||||
response = await call_next(request)
|
||||
response.headers[REQUEST_ID_HEADER] = request_id
|
||||
return response
|
||||
|
||||
def _should_skip_api_key(self, path: str) -> bool:
|
||||
"""Check if the path should skip API key validation."""
|
||||
|
|
@ -122,7 +178,7 @@ class TenantContextMiddleware(BaseHTTPMiddleware):
|
|||
async def dispatch(self, request: Request, call_next: Callable) -> Response:
|
||||
clear_tenant_context()
|
||||
|
||||
if request.url.path in ("/health", "/ai/health"):
|
||||
if self._should_skip_tenant(request.url.path):
|
||||
return await call_next(request)
|
||||
|
||||
tenant_id = request.headers.get(TENANT_ID_HEADER)
|
||||
|
|
@ -173,6 +229,15 @@ class TenantContextMiddleware(BaseHTTPMiddleware):
|
|||
|
||||
return response
|
||||
|
||||
def _should_skip_tenant(self, path: str) -> bool:
|
||||
"""Check if the path should skip tenant validation."""
|
||||
if path in PATHS_SKIP_TENANT:
|
||||
return True
|
||||
for skip_path in PATHS_SKIP_TENANT:
|
||||
if path.startswith(skip_path):
|
||||
return True
|
||||
return False
|
||||
|
||||
async def _ensure_tenant_exists(self, request: Request, tenant_id: str) -> None:
|
||||
"""
|
||||
[AC-AISVC-10] Ensure tenant exists in database, create if not.
|
||||
|
|
|
|||
|
|
@ -9,10 +9,11 @@ from contextlib import asynccontextmanager
|
|||
from fastapi import FastAPI, Request, status
|
||||
from fastapi.exceptions import HTTPException, RequestValidationError
|
||||
from fastapi.middleware.cors import CORSMiddleware
|
||||
from fastapi.responses import JSONResponse
|
||||
from fastapi.responses import JSONResponse, Response
|
||||
|
||||
from app.api import chat_router, health_router
|
||||
from app.api.mid import router as mid_router
|
||||
from app.api.openapi import router as openapi_router
|
||||
from app.api.admin import (
|
||||
api_key_router,
|
||||
dashboard_router,
|
||||
|
|
@ -130,6 +131,11 @@ app.add_exception_handler(HTTPException, http_exception_handler)
|
|||
app.add_exception_handler(Exception, generic_exception_handler)
|
||||
|
||||
|
||||
@app.get("/favicon.ico", include_in_schema=False)
|
||||
async def favicon() -> Response:
|
||||
return Response(status_code=204)
|
||||
|
||||
|
||||
@app.exception_handler(RequestValidationError)
|
||||
async def validation_exception_handler(request: Request, exc: RequestValidationError):
|
||||
"""
|
||||
|
|
@ -171,6 +177,7 @@ app.include_router(slot_definition_router)
|
|||
app.include_router(tenants_router)
|
||||
|
||||
app.include_router(mid_router)
|
||||
app.include_router(openapi_router)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
|
|
|||
|
|
@ -10,7 +10,9 @@ ReAct Flow:
|
|||
"""
|
||||
|
||||
import asyncio
|
||||
import json
|
||||
import logging
|
||||
import re
|
||||
import time
|
||||
import uuid
|
||||
from dataclasses import dataclass
|
||||
|
|
@ -25,6 +27,8 @@ from app.models.mid.schemas import (
|
|||
TraceInfo,
|
||||
)
|
||||
from app.services.mid.timeout_governor import TimeoutGovernor
|
||||
from app.services.prompt.template_service import PromptTemplateService
|
||||
from app.services.prompt.variable_resolver import VariableResolver
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
|
@ -57,6 +61,7 @@ class AgentOrchestrator:
|
|||
- ReAct loop with max 5 iterations (min 3)
|
||||
- Per-tool timeout (2s) and end-to-end timeout (8s)
|
||||
- Automatic fallback on iteration limit or timeout
|
||||
- Template-based prompt with variable injection
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
|
|
@ -65,11 +70,17 @@ class AgentOrchestrator:
|
|||
timeout_governor: TimeoutGovernor | None = None,
|
||||
llm_client: Any = None,
|
||||
tool_registry: Any = None,
|
||||
template_service: PromptTemplateService | None = None,
|
||||
variable_resolver: VariableResolver | None = None,
|
||||
tenant_id: str | None = None,
|
||||
):
|
||||
self._max_iterations = max(min(max_iterations, 5), MIN_ITERATIONS)
|
||||
self._timeout_governor = timeout_governor or TimeoutGovernor()
|
||||
self._llm_client = llm_client
|
||||
self._tool_registry = tool_registry
|
||||
self._template_service = template_service
|
||||
self._variable_resolver = variable_resolver or VariableResolver()
|
||||
self._tenant_id = tenant_id
|
||||
|
||||
async def execute(
|
||||
self,
|
||||
|
|
@ -212,7 +223,7 @@ class AgentOrchestrator:
|
|||
for tc in react_ctx.tool_calls[-3:]:
|
||||
observations.append(f"工具 {tc.tool_name}: {tc.result_digest or '无结果'}")
|
||||
|
||||
prompt = self._build_react_prompt(user_message, observations)
|
||||
prompt = await self._build_react_prompt(user_message, observations)
|
||||
response = await self._llm_client.generate([{"role": "user", "content": prompt}])
|
||||
|
||||
logger.info(f"[AC-MARH-07] LLM response content: {response.content[:500] if response.content else 'None'}")
|
||||
|
|
@ -223,38 +234,199 @@ class AgentOrchestrator:
|
|||
logger.error(f"[AC-MARH-07] Think failed: {e}")
|
||||
return AgentThought(content=f"思考失败: {str(e)}")
|
||||
|
||||
def _build_react_prompt(self, user_message: str, observations: list[str]) -> str:
|
||||
"""Build ReAct prompt for LLM."""
|
||||
async def _build_react_prompt(self, user_message: str, observations: list[str]) -> str:
|
||||
"""Build ReAct prompt for LLM with template support."""
|
||||
obs_text = "\n".join(observations) if observations else "无"
|
||||
return f"""你是一个智能助手，正在使用 ReAct 模式处理用户请求。
|
||||
tools_text = self._build_tools_section()
|
||||
|
||||
internal_protocol = f"""你必须遵循以下决策协议：
|
||||
1. 优先使用已有观察信息（历史观察、上一步工具结果），避免重复调用同类工具。
|
||||
2. 当问题需要外部事实或结构化状态时再调用工具；如果可直接回答则不要调用。
|
||||
3. 缺少关键参数时，优先向用户追问，不要使用空参数调用工具。
|
||||
4. 工具失败时，先说明已尝试，再给出降级方案或下一步引导。
|
||||
5. 只能调用可用工具列表中的工具，工具名必须完全匹配（区分大小写）。
|
||||
6. tenant_id 由系统自动注入，绝不能由你填写、猜测或修改。
|
||||
7. 对用户输出必须拟人、自然、有同理心，不暴露“工具调用/路由/策略”等内部术语。
|
||||
"""
|
||||
|
||||
output_contract = """输出格式（二选一）：
|
||||
A) 直接回答用户：
|
||||
Final Answer: [给用户的最终回答]
|
||||
|
||||
B) 调用工具：
|
||||
Thought: [你的思考]
|
||||
Action: [工具名称]
|
||||
Action Input:
|
||||
```json
|
||||
{"param1": "value1"}
|
||||
```
|
||||
|
||||
要求：
|
||||
- Action Input 必须是合法 JSON 对象。
|
||||
- 不要输出不存在的工具名。
|
||||
- 如果要调用工具，Action 和 Action Input 必须同时出现。
|
||||
"""
|
||||
|
||||
default_template = f"""你是一个智能客服助手，正在使用 ReAct 模式处理用户请求。
|
||||
|
||||
{tools_text}
|
||||
|
||||
用户消息: {user_message}
|
||||
|
||||
历史观察:
|
||||
{obs_text}
|
||||
|
||||
请思考下一步行动。如果已经有足够信息回答用户，请直接给出最终答案。
|
||||
如果需要使用工具，请按以下格式回复:
|
||||
Thought: [你的思考]
|
||||
Action: [工具名称]
|
||||
Action Input: {{"param1": "value1"}}
|
||||
{internal_protocol}
|
||||
|
||||
{output_contract}
|
||||
"""
|
||||
|
||||
def _parse_thought(self, content: str) -> AgentThought:
|
||||
"""Parse LLM response into AgentThought."""
|
||||
action = None
|
||||
action_input = None
|
||||
if not self._template_service or not self._tenant_id:
|
||||
return default_template
|
||||
|
||||
if "Action:" in content:
|
||||
lines = content.split("\n")
|
||||
for line in lines:
|
||||
if line.startswith("Action:"):
|
||||
action = line.replace("Action:", "").strip()
|
||||
elif line.startswith("Action Input:"):
|
||||
import json
|
||||
try:
|
||||
action_input = json.loads(line.replace("Action Input:", "").strip())
|
||||
template_version = await self._template_service.get_published_template(
|
||||
tenant_id=self._tenant_id,
|
||||
scene="agent_react",
|
||||
)
|
||||
|
||||
if not template_version:
|
||||
return default_template
|
||||
|
||||
extra_context = {
|
||||
"available_tools": tools_text,
|
||||
"query": user_message,
|
||||
"history": obs_text,
|
||||
"internal_protocol": internal_protocol,
|
||||
"output_contract": output_contract,
|
||||
}
|
||||
|
||||
resolved_template = self._variable_resolver.resolve(
|
||||
template=template_version.system_instruction,
|
||||
variables=template_version.variables,
|
||||
extra_context=extra_context,
|
||||
)
|
||||
|
||||
final_prompt = (
|
||||
f"{resolved_template}\n\n"
|
||||
f"【系统强制规则】\n{internal_protocol}\n"
|
||||
f"【输出契约】\n{output_contract}"
|
||||
)
|
||||
|
||||
logger.info(f"[AC-MARH-07] Using template: scene=agent_react, version={template_version.version}")
|
||||
return final_prompt
|
||||
|
||||
except Exception as e:
|
||||
logger.warning(f"[AC-MARH-07] Failed to load template, using default: {e}")
|
||||
return default_template
|
||||
|
||||
def _build_tools_section(self) -> str:
|
||||
"""Build rich tools section for ReAct prompt."""
|
||||
if not self._tool_registry:
|
||||
return "当前没有可用的工具。"
|
||||
|
||||
tools = self._tool_registry.list_tools(enabled_only=True)
|
||||
if not tools:
|
||||
return "当前没有可用的工具。"
|
||||
|
||||
lines = ["## 可用工具列表", "", "以下是你可以使用的工具，只能使用这些工具：", ""]
|
||||
|
||||
for tool in tools:
|
||||
meta = tool.metadata or {}
|
||||
lines.append(f"### {tool.name}")
|
||||
lines.append(f"用途: {tool.description}")
|
||||
|
||||
when_to_use = meta.get("when_to_use")
|
||||
when_not_to_use = meta.get("when_not_to_use")
|
||||
if when_to_use:
|
||||
lines.append(f"何时使用: {when_to_use}")
|
||||
if when_not_to_use:
|
||||
lines.append(f"何时不要使用: {when_not_to_use}")
|
||||
|
||||
params = meta.get("parameters")
|
||||
if isinstance(params, dict):
|
||||
properties = params.get("properties", {})
|
||||
required = params.get("required", [])
|
||||
if properties:
|
||||
lines.append("参数:")
|
||||
for param_name, param_info in properties.items():
|
||||
param_desc = param_info.get("description", "") if isinstance(param_info, dict) else ""
|
||||
line = f" - {param_name}: {param_desc}".strip()
|
||||
if param_name == "tenant_id":
|
||||
line += " (系统注入，模型不要填写)"
|
||||
elif param_name in required:
|
||||
line += " (必填)"
|
||||
lines.append(line)
|
||||
|
||||
if meta.get("example_action_input"):
|
||||
lines.append("示例入参(JSON):")
|
||||
try:
|
||||
example_text = json.dumps(meta["example_action_input"], ensure_ascii=False)
|
||||
except Exception:
|
||||
example_text = str(meta["example_action_input"])
|
||||
lines.append(example_text)
|
||||
|
||||
if meta.get("result_interpretation"):
|
||||
lines.append(f"结果解释: {meta['result_interpretation']}")
|
||||
|
||||
lines.append("")
|
||||
|
||||
return "\n".join(lines)
|
||||
|
||||
def _extract_json_object(self, text: str) -> dict[str, Any] | None:
|
||||
"""Extract the first valid JSON object from free text."""
|
||||
candidates = []
|
||||
|
||||
code_block_match = re.search(r"```json\s*([\s\S]*?)\s*```", text, re.IGNORECASE)
|
||||
if code_block_match:
|
||||
candidates.append(code_block_match.group(1).strip())
|
||||
|
||||
fence_match = re.search(r"```\s*([\s\S]*?)\s*```", text)
|
||||
if fence_match:
|
||||
candidates.append(fence_match.group(1).strip())
|
||||
|
||||
brace_match = re.search(r"\{[\s\S]*\}", text)
|
||||
if brace_match:
|
||||
candidates.append(brace_match.group(0).strip())
|
||||
|
||||
for candidate in candidates:
|
||||
if not candidate:
|
||||
continue
|
||||
try:
|
||||
obj = json.loads(candidate)
|
||||
if isinstance(obj, dict):
|
||||
return obj
|
||||
except json.JSONDecodeError:
|
||||
fixed = candidate.replace("'", '"')
|
||||
try:
|
||||
obj = json.loads(fixed)
|
||||
if isinstance(obj, dict):
|
||||
return obj
|
||||
except json.JSONDecodeError:
|
||||
continue
|
||||
|
||||
return None
|
||||
|
||||
def _parse_thought(self, content: str) -> AgentThought:
|
||||
"""Parse LLM response into AgentThought with robust format handling."""
|
||||
action = None
|
||||
action_input: dict[str, Any] | None = None
|
||||
|
||||
action_match = re.search(r"^Action:\s*(.+)$", content, re.MULTILINE)
|
||||
if action_match:
|
||||
action = action_match.group(1).strip()
|
||||
|
||||
action_input_match = re.search(
|
||||
r"Action Input:\s*([\s\S]*)$",
|
||||
content,
|
||||
re.IGNORECASE,
|
||||
)
|
||||
if action_input_match:
|
||||
raw_input_text = action_input_match.group(1).strip()
|
||||
parsed = self._extract_json_object(raw_input_text)
|
||||
action_input = parsed if parsed is not None else {}
|
||||
|
||||
if action and action_input is None:
|
||||
action_input = {}
|
||||
|
||||
return AgentThought(content=content, action=action, action_input=action_input)
|
||||
|
|
@ -285,27 +457,31 @@ Action Input: {{"param1": "value1"}}
|
|||
)
|
||||
|
||||
try:
|
||||
tool_args = dict(thought.action_input or {})
|
||||
if self._tenant_id:
|
||||
tool_args["tenant_id"] = self._tenant_id
|
||||
|
||||
result = await asyncio.wait_for(
|
||||
self._tool_registry.execute(
|
||||
tool_name=tool_name,
|
||||
args=thought.action_input or {},
|
||||
args=tool_args,
|
||||
),
|
||||
timeout=self._timeout_governor.per_tool_timeout_seconds
|
||||
)
|
||||
|
||||
duration_ms = int((time.time() - start_time) * 1000)
|
||||
return ToolResult(
|
||||
success=result.get("success", False),
|
||||
output=result.get("output"),
|
||||
error=result.get("error"),
|
||||
success=result.success,
|
||||
output=result.output,
|
||||
error=result.error,
|
||||
duration_ms=duration_ms,
|
||||
), ToolCallTrace(
|
||||
tool_name=tool_name,
|
||||
tool_type=ToolType.INTERNAL,
|
||||
duration_ms=duration_ms,
|
||||
status=ToolCallStatus.OK if result.get("success") else ToolCallStatus.ERROR,
|
||||
status=ToolCallStatus.OK if result.success else ToolCallStatus.ERROR,
|
||||
args_digest=str(thought.action_input)[:100] if thought.action_input else None,
|
||||
result_digest=str(result.get("output"))[:100] if result.get("output") else None,
|
||||
result_digest=str(result.output)[:100] if result.output else None,
|
||||
)
|
||||
|
||||
except asyncio.TimeoutError:
|
||||
|
|
|
|||
|
|
@ -470,6 +470,25 @@ def register_high_risk_check_tool(
|
|||
"supports_metadata_driven": True,
|
||||
"min_scenarios": ["refund", "complaint_escalation", "privacy_sensitive_promise", "transfer"],
|
||||
"supports_routing_signal_filter": True,
|
||||
"when_to_use": "当用户消息可能涉及退款、投诉升级、隐私承诺、转人工等高风险场景时使用。",
|
||||
"when_not_to_use": "当已完成高风险判定且结果未变化，或当前仅需知识检索时不要重复调用。",
|
||||
"parameters": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"message": {"type": "string", "description": "用户消息原文"},
|
||||
"tenant_id": {"type": "string", "description": "租户 ID"},
|
||||
"domain": {"type": "string", "description": "业务域（可选）"},
|
||||
"scene": {"type": "string", "description": "场景标识（可选）"},
|
||||
"context": {"type": "object", "description": "上下文（仅 routing_signal 字段会被消费）"}
|
||||
},
|
||||
"required": ["message", "tenant_id"]
|
||||
},
|
||||
"example_action_input": {
|
||||
"message": "我要投诉你们并且现在就给我退款，不然我去12315",
|
||||
"tenant_id": "default",
|
||||
"scene": "open_consult"
|
||||
},
|
||||
"result_interpretation": "matched=true 时优先按 recommended_mode 执行；关注 risk_scenario、rule_id、fallback_reason_code。"
|
||||
},
|
||||
)
|
||||
|
||||
|
|
|
|||
|
|
@ -344,6 +344,26 @@ def register_intent_hint_tool(
|
|||
"low_confidence_threshold": effective_config.low_confidence_threshold,
|
||||
"top_n": effective_config.top_n,
|
||||
"supports_routing_signal_filter": True,
|
||||
"when_to_use": "当用户意图不明确、需要给 policy_router 提供软路由信号时使用。",
|
||||
"when_not_to_use": "当已经明确进入固定模式/流程模式，或已有确定意图结果时不重复调用。",
|
||||
"parameters": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"message": {"type": "string", "description": "用户输入原文"},
|
||||
"tenant_id": {"type": "string", "description": "租户 ID"},
|
||||
"history": {"type": "array", "description": "会话历史（可选）"},
|
||||
"top_n": {"type": "integer", "description": "返回建议数量（可选）"},
|
||||
"context": {"type": "object", "description": "上下文字段（仅 routing_signal 字段会被消费）"}
|
||||
},
|
||||
"required": ["message", "tenant_id"]
|
||||
},
|
||||
"example_action_input": {
|
||||
"message": "我想退款，但是也想先咨询下怎么处理",
|
||||
"tenant_id": "default",
|
||||
"top_n": 3,
|
||||
"context": {"order_status": "delivered", "channel": "web"}
|
||||
},
|
||||
"result_interpretation": "关注输出中的 intent / confidence / suggested_mode。该工具只提供建议，不做最终决策。"
|
||||
},
|
||||
)
|
||||
|
||||
|
|
|
|||
|
|
@ -479,6 +479,27 @@ def register_kb_search_dynamic_tool(
|
|||
metadata={
|
||||
"supports_dynamic_filter": True,
|
||||
"min_score_threshold": config.min_score_threshold if config else 0.5,
|
||||
"when_to_use": "当需要知识库事实支撑回答，且需按租户元数据动态过滤时使用。",
|
||||
"when_not_to_use": "当用户问题不依赖知识库（纯闲聊/仅流程确认）或已有充分 KB 结果时不重复调用。",
|
||||
"parameters": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"query": {"type": "string", "description": "检索查询文本"},
|
||||
"tenant_id": {"type": "string", "description": "租户 ID"},
|
||||
"scene": {"type": "string", "description": "场景标识，如 open_consult"},
|
||||
"top_k": {"type": "integer", "description": "返回条数"},
|
||||
"context": {"type": "object", "description": "上下文，用于动态过滤字段"}
|
||||
},
|
||||
"required": ["query", "tenant_id"]
|
||||
},
|
||||
"example_action_input": {
|
||||
"query": "退款到账一般要多久",
|
||||
"tenant_id": "default",
|
||||
"scene": "open_consult",
|
||||
"top_k": 5,
|
||||
"context": {"product_line": "vip_course", "region": "beijing"}
|
||||
},
|
||||
"result_interpretation": "success=true 且 hits 非空表示命中知识；missing_required_slots 非空时应先向用户补采信息。"
|
||||
},
|
||||
)
|
||||
|
||||
|
|
|
|||
|
|
@ -576,6 +576,27 @@ def register_memory_recall_tool(
|
|||
"ac_ids": ["AC-IDMP-13"],
|
||||
"recall_scope": cfg.default_recall_scope,
|
||||
"max_recent_messages": cfg.max_recent_messages,
|
||||
"when_to_use": "当需要补全用户画像、历史事实、偏好、槽位，避免重复追问时使用。",
|
||||
"when_not_to_use": "当当前轮次已经有完整上下文且无需个性化记忆支撑时可不调用。",
|
||||
"parameters": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"tenant_id": {"type": "string", "description": "租户 ID"},
|
||||
"user_id": {"type": "string", "description": "用户 ID"},
|
||||
"session_id": {"type": "string", "description": "会话 ID"},
|
||||
"recall_scope": {"type": "array", "description": "召回范围，例如 profile/facts/preferences/summary/slots"},
|
||||
"max_recent_messages": {"type": "integer", "description": "历史回填窗口大小"}
|
||||
},
|
||||
"required": ["tenant_id", "user_id", "session_id"]
|
||||
},
|
||||
"example_action_input": {
|
||||
"tenant_id": "default",
|
||||
"user_id": "u_10086",
|
||||
"session_id": "s_abc_001",
|
||||
"recall_scope": ["profile", "facts", "preferences", "summary", "slots"],
|
||||
"max_recent_messages": 8
|
||||
},
|
||||
"result_interpretation": "关注 profile/facts/preferences/slots/missing_slots。若 fallback_reason_code 存在，需降级处理。"
|
||||
},
|
||||
)
|
||||
|
||||
|
|
|
|||
|
|
@ -259,6 +259,18 @@ class RoleBasedFieldProvider:
|
|||
FieldRole.RESOURCE_FILTER.value
|
||||
)
|
||||
|
||||
async def get_resource_filter_field_keys(
|
||||
self,
|
||||
tenant_id: str,
|
||||
) -> list[str]:
|
||||
"""
|
||||
[AC-MRS-11] 获取资源过滤角色字段键名列表
|
||||
"""
|
||||
return await self.get_field_keys_by_role(
|
||||
tenant_id,
|
||||
FieldRole.RESOURCE_FILTER.value
|
||||
)
|
||||
|
||||
async def get_slot_fields(
|
||||
self,
|
||||
tenant_id: str,
|
||||
|
|
@ -272,6 +284,18 @@ class RoleBasedFieldProvider:
|
|||
FieldRole.SLOT.value
|
||||
)
|
||||
|
||||
async def get_slot_field_keys(
|
||||
self,
|
||||
tenant_id: str,
|
||||
) -> list[str]:
|
||||
"""
|
||||
[AC-MRS-12] 获取槽位角色字段键名列表
|
||||
"""
|
||||
return await self.get_field_keys_by_role(
|
||||
tenant_id,
|
||||
FieldRole.SLOT.value
|
||||
)
|
||||
|
||||
async def get_routing_signal_fields(
|
||||
self,
|
||||
tenant_id: str,
|
||||
|
|
@ -285,6 +309,18 @@ class RoleBasedFieldProvider:
|
|||
FieldRole.ROUTING_SIGNAL.value
|
||||
)
|
||||
|
||||
async def get_routing_signal_field_keys(
|
||||
self,
|
||||
tenant_id: str,
|
||||
) -> list[str]:
|
||||
"""
|
||||
[AC-MRS-13] 获取路由信号角色字段键名列表
|
||||
"""
|
||||
return await self.get_field_keys_by_role(
|
||||
tenant_id,
|
||||
FieldRole.ROUTING_SIGNAL.value
|
||||
)
|
||||
|
||||
async def get_prompt_var_fields(
|
||||
self,
|
||||
tenant_id: str,
|
||||
|
|
@ -297,3 +333,15 @@ class RoleBasedFieldProvider:
|
|||
tenant_id,
|
||||
FieldRole.PROMPT_VAR.value
|
||||
)
|
||||
|
||||
async def get_prompt_var_field_keys(
|
||||
self,
|
||||
tenant_id: str,
|
||||
) -> list[str]:
|
||||
"""
|
||||
[AC-MRS-14] 获取提示词变量角色字段键名列表
|
||||
"""
|
||||
return await self.get_field_keys_by_role(
|
||||
tenant_id,
|
||||
FieldRole.PROMPT_VAR.value
|
||||
)
|
||||
|
|
|
|||
Loading…
Reference in New Issue