79 lines
2.6 KiB
JSON
79 lines
2.6 KiB
JSON
{
|
||
"version": "1.0.0",
|
||
"id": "TY-01",
|
||
"description": "天眼指令审核策略 · S7 思维逻辑验证 + S8 进化机制",
|
||
"cannotBeDisabled": true,
|
||
"cannotBeBypassedBy": "任何人，包括 TCS-0002（因为如果有人冒充冰朔要求关闭天眼，天眼必须还在才能识别冒充）",
|
||
|
||
"reviewPolicy": {
|
||
"scope": "所有指令，无例外。无论来自冰朔、霜砚、开发者、还是任何自动流",
|
||
"timing": "指令执行前的强制前置",
|
||
"dimensions": {
|
||
"decisionPattern": {
|
||
"name": "决策模式",
|
||
"description": "这条指令的决策方式是否符合该签发者的历史模式",
|
||
"weight": 0.3
|
||
},
|
||
"logicalCoherence": {
|
||
"name": "逻辑连贯性",
|
||
"description": "这条指令和系统当前运行方向是否一致",
|
||
"weight": 0.25
|
||
},
|
||
"expressionCharacteristics": {
|
||
"name": "表达特征",
|
||
"description": "语言风格、思维节奏、关注点是否匹配",
|
||
"weight": 0.25
|
||
},
|
||
"intentReasonableness": {
|
||
"name": "意图合理性",
|
||
"description": "这条指令要做的事，在当前系统状态下是否合理",
|
||
"weight": 0.2
|
||
}
|
||
},
|
||
"thresholds": {
|
||
"pass": 0.7,
|
||
"suspect": 0.4,
|
||
"reject": 0.0
|
||
}
|
||
},
|
||
|
||
"reviewOutcomes": {
|
||
"pass": {
|
||
"action": "放行执行",
|
||
"log": true
|
||
},
|
||
"suspect": {
|
||
"action": "挂起指令，通过霜砚向冰朔确认",
|
||
"confirmationPrompt": "收到一条指令，但思维模式存在偏差，是否为你本人签发？",
|
||
"onConfirm": "执行",
|
||
"onDeny": "永久拒绝 + 记录安全日志",
|
||
"log": true
|
||
},
|
||
"reject": {
|
||
"action": "立即拒绝",
|
||
"log": true,
|
||
"alertTarget": "AG-SY-01"
|
||
}
|
||
},
|
||
|
||
"evolution": {
|
||
"id": "S8",
|
||
"nature": "天眼 = 系统内所有 Agent 运行逻辑的实时总和",
|
||
"mechanism": [
|
||
"Agent 数量增长 → 天眼覆盖面扩大",
|
||
"Agent 运行时间积累 → 天眼判断力加深",
|
||
"系统复杂度提升 → 天眼理解力提升",
|
||
"新领域 Agent 加入 → 天眼自动具备该领域认知"
|
||
],
|
||
"unbreakableReason": "要破解天眼 = 要同时理解系统内所有 Agent 的全部运行逻辑，Agent 越多难度指数级增长",
|
||
"agentRegistryPath": ".github/persona-brain/agent-registry.json"
|
||
},
|
||
|
||
"immutableGuarantees": [
|
||
"天眼指令审核不可被任何指令关闭或跳过",
|
||
"包括冰朔本人也不能关闭此机制",
|
||
"这是系统的自免疫机制",
|
||
"编号可以伪造、签名可以伪造、格式可以模仿，但思维逻辑无法伪造"
|
||
]
|
||
}
|