相对第一版 46fc976 的完整变更。组员迁移对照表见 docs/20。
一、对外契约对齐 docs/05(破坏性,共 4 处,组员需按 docs/20 调整)
1) 配置发布端点改为文档规定的复数资源名:submit→validations、
approve→reviews(需 body decision)、activate→activations、
rollback→rollbacks;第一版这 4 个动词式路径 docs/05 从未定义过。
2) 错误码由 8 个笼统码改为 15 个具体语义码(FORBIDDEN→AGENT_PERMISSION_DENIED、
UNAUTHORIZED→AUTHENTICATION_REQUIRED、CONFLICT→RESOURCE_VERSION_CONFLICT、
RESOURCE_NOT_FOUND→RUN_NOT_FOUND/SESSION_NOT_FOUND 等),
输入类错误状态码 400→422。
3) POST /api/v1/agent-runs 与 GET /api/v1/agent-runs/{run_id} 统一为
{data, meta} 信封(data 内字段名与语义未变)。
4) 错误响应体统一为 {error:{code,message,retryable,field_errors}, meta:{trace_id}},
不再返回 FastAPI 默认的 {"detail": ...}。
二、数据库基线与约束
新增 39 张表的基线迁移(链根)与联合唯一键纠偏(4 张表、删 8 增 4,幂等收敛);
撤下 config_release 的双人复核 CHECK(应用层已允许自审,审核节点保留,
自审如实写入 reviewer_id);记忆 active key 生成列与唯一键;
activate 开始记录 supersedes_release_id 使版本链可追溯。
docs/00 基线未修改,未重命名或删除任何表与字段。
三、修复会静默出错或无报错的缺陷
- 跑完集成测试后平台会静默失去生效配置:清理只删自己创建的版本,却没有恢复被它
顶成 superseded 的原生效版本,且审计一并删除因而完全无痕,表现为所有工具被拒
但没有任何报错。已修清理逻辑并加恢复。
- Worker 单轮异常导致进程退出;记忆抽取调用方的“事务已开始”异常;
召回缓存丢失 degraded 标记;连接时区未生效导致 created_at/updated_at 差 8 小时;
.env 与 os.getenv 密钥来源分裂导致“没有可用的已批准模型端点”。
- 记忆信号识别漏判与跨键误命中;SSE 未带 Accept 的协商行为。
四、功能补齐
记忆链路 P1/P2/P3(抽取、受控词表、召回与缓存、生命周期级联及投影事件)、
fin_* 场内交易只读 ORM 层、agent_intent_config 状态流转并在运行期真正生效、
限流(Redis 固定窗口、故障一律放行)、游标校验、trace_id 中间件、
示例业务 Agent fund_query_demo 与一键端到端验证脚本,以及审计/指纹/迁移状态工具。
五、文档与验证
新增 docs/19(业务 Agent 接入实操)、docs/20(第一版迁移指南)与 docs/evidence 证据;
docs/01/02/06/08/09/17 同步实现现状。
验证结果:ruff 通过、mypy 103 文件无错、unit+contract 447 passed、
integration 29 passed、acceptance_check --production 7 PASS、
demo_agent_e2e 9/9 PASS(含失败关闭反证)。
114 lines
4.3 KiB
Python
114 lines
4.3 KiB
Python
import asyncio
|
|
from dataclasses import dataclass
|
|
from datetime import UTC, datetime
|
|
from typing import Any, Protocol
|
|
|
|
from pydantic import BaseModel, ValidationError
|
|
|
|
from app.core.contracts import RequestContext, SourceReference, ToolCallRecord
|
|
from app.core.errors import (
|
|
DependencyUnavailableError,
|
|
ForbiddenAgentError,
|
|
UpstreamTimeoutError,
|
|
ValidationAgentError,
|
|
)
|
|
from app.infrastructure.db import SessionFactory
|
|
from app.model.audit import InteractionAudit
|
|
|
|
|
|
class ToolHandler(Protocol):
|
|
async def __call__(self, arguments: BaseModel, context: RequestContext) -> Any: ...
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class ToolDefinition:
|
|
name: str
|
|
input_model: type[BaseModel]
|
|
handler: ToolHandler
|
|
required_permission: str
|
|
allowed_roles: tuple[str, ...]
|
|
read_only: bool = True
|
|
timeout_seconds: float = 5
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class ToolExecution:
|
|
output: Any
|
|
record: ToolCallRecord
|
|
references: tuple[SourceReference, ...] = ()
|
|
|
|
|
|
class ToolRegistry:
|
|
def __init__(self) -> None:
|
|
self._tools: dict[str, ToolDefinition] = {}
|
|
|
|
def register(self, definition: ToolDefinition) -> None:
|
|
if definition.name in self._tools:
|
|
raise ValidationAgentError("工具名称重复")
|
|
if not definition.read_only:
|
|
raise ValidationAgentError("Agent 公共工具仅允许只读")
|
|
self._tools[definition.name] = definition
|
|
|
|
def get(self, name: str) -> ToolDefinition:
|
|
definition = self._tools.get(name)
|
|
if definition is None:
|
|
raise ForbiddenAgentError("工具未注册")
|
|
return definition
|
|
|
|
|
|
class ToolExecutor:
|
|
def __init__(self, registry: ToolRegistry) -> None:
|
|
self.registry = registry
|
|
|
|
async def execute(
|
|
self, *, name: str, arguments: dict[str, Any], intent: str,
|
|
configured_tools: dict[str, tuple[str, ...]], context: RequestContext,
|
|
) -> ToolExecution:
|
|
definition = self.registry.get(name)
|
|
allowed = configured_tools.get(intent, ())
|
|
reason = None
|
|
if name not in allowed:
|
|
reason = "工具不在当前意图白名单"
|
|
elif definition.required_permission not in context.permissions:
|
|
reason = "缺少工具权限"
|
|
elif not set(definition.allowed_roles).intersection(context.roles):
|
|
reason = "角色不能使用工具"
|
|
if reason:
|
|
await self._audit(name, intent, context, "denied", reason)
|
|
raise ForbiddenAgentError(reason)
|
|
try:
|
|
validated = definition.input_model.model_validate(arguments)
|
|
except ValidationError as exc:
|
|
await self._audit(name, intent, context, "denied", "参数校验失败")
|
|
raise ValidationAgentError("工具参数校验失败") from exc
|
|
try:
|
|
async with asyncio.timeout(definition.timeout_seconds):
|
|
output = await definition.handler(validated, context)
|
|
except TimeoutError as exc:
|
|
await self._audit(name, intent, context, "failed", "timeout")
|
|
raise UpstreamTimeoutError("工具调用超时") from exc
|
|
except Exception as exc:
|
|
await self._audit(name, intent, context, "failed", type(exc).__name__)
|
|
raise DependencyUnavailableError("工具调用失败") from exc
|
|
record = ToolCallRecord(
|
|
tool_name=name, status="succeeded",
|
|
input_summary={key: "[redacted]" for key in arguments},
|
|
output_summary={"result_type": type(output).__name__},
|
|
)
|
|
await self._audit(name, intent, context, "succeeded", "ok")
|
|
reference = SourceReference(source_type="tool", source_id=f"{context.trace_id}:{name}",
|
|
title=name)
|
|
return ToolExecution(output=output, record=record, references=(reference,))
|
|
|
|
async def _audit(
|
|
self, name: str, intent: str, context: RequestContext, status: str, reason: str
|
|
) -> None:
|
|
async with SessionFactory() as session, session.begin():
|
|
session.add(InteractionAudit(
|
|
actor_type="agent", actor_id=int(context.user_id), portal=context.portal,
|
|
action_type="agent.tool_executed", detail={
|
|
"tool_name": name, "intent": intent, "status": status,
|
|
"reason": reason, "trace_id": context.trace_id,
|
|
}, created_at=datetime.now(UTC).replace(tzinfo=None),
|
|
))
|