Policies：全部命中与首个恢复

策略集收集所有匹配理由；恢复链按顺序选择第一个显式 handled 结果。

assess(input) 返回 {required,reasons,kinds}，不会只取第一个匹配，也不会让某个“快速模式”覆盖强制规则。RecoveryResult 是 {handled:false} 或 {handled:true,value}；值为 undefined 仍然可以是已处理。未知 ID、重复 ID、空描述和异步 handler 均拒绝。具体 kind 与恢复条件由 App 定义。

由简到繁的同业务示例

为同一答复声明何时必须复核

入门 · Node.js

TypeScript

import { Policies } from '@jangts/agent-core/kits/policies';
const policy = Policies.definePolicySet([
  { id: 'has-write', description: '有写操作就复核', kind: 'review', when: (input: { wrote: boolean }) => input.wrote },
]);
console.log(policy.assess({ wrote: true }));

JavaScript

import { Policies } from '@jangts/agent-core/kits/policies';
const policy = Policies.definePolicySet([
    { id: 'has-write', description: '有写操作就复核', kind: 'review', when: (input) => input.wrote },
]);
console.log(policy.assess({ wrote: true }));

增加另一理由及有序恢复

配置 · Node.js

TypeScript

import { Policies } from '@jangts/agent-core/kits/policies';
type Input = { wrote: boolean; uncertain: boolean; cancelled: boolean };
const policy = Policies.definePolicySet<Input, 'review'>([
  { id: 'has-write', description: '有写操作就复核', kind: 'review', when: input => input.wrote },
  { id: 'uncertain', description: '证据不确定就复核', kind: 'review', when: input => input.uncertain },
]);
const recovery = Policies.defineRecoveryChain<Input, string>([
  { id: 'cancelled', description: '取消时不发布答案', tryRecover: input => input.cancelled ? { handled: true, value: '任务已取消' } : { handled: false } },
  { id: 'needs-review', description: '未通过复核时明确说明', tryRecover: input => policy.assess(input).required ? { handled: true, value: '结果还需要复核' } : { handled: false } },
]);
const input = { wrote: true, uncertain: true, cancelled: false };
console.log(policy.assess(input), recovery.tryRecover(input));

JavaScript

import { Policies } from '@jangts/agent-core/kits/policies';
const policy = Policies.definePolicySet([
    { id: 'has-write', description: '有写操作就复核', kind: 'review', when: input => input.wrote },
    { id: 'uncertain', description: '证据不确定就复核', kind: 'review', when: input => input.uncertain },
]);
const recovery = Policies.defineRecoveryChain([
    { id: 'cancelled', description: '取消时不发布答案', tryRecover: input => input.cancelled ? { handled: true, value: '任务已取消' } : { handled: false } },
    { id: 'needs-review', description: '未通过复核时明确说明', tryRecover: input => policy.assess(input).required ? { handled: true, value: '结果还需要复核' } : { handled: false } },
]);
const input = { wrote: true, uncertain: true, cancelled: false };
console.log(policy.assess(input), recovery.tryRecover(input));

策略与实现

• 全命中与首个 handled