{"item":{"id":"obs-topic-approvals","product":"muse","language":"zh","title":"agent 做哪些事之前应该先问？","summary":"围绕具体动作、接收方和额度定义审批。对话中的承诺不能单独证明执行边界。","contentType":"guide","category":"security","revision":2,"canonicalUrl":"https://botclaw.tech/items/obs-topic-approvals?lang=zh","sourceUrls":["https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse","https://docs.openclaw.ai/gateway/security","https://hermes-agent.nousresearch.com/docs/user-guide/security","https://help.openai.com/en/articles/20001530-getting-started-with-your-dot"],"sourcePublishedAt":"2026-10-01T16:24:26.193Z","reviewedAt":"2026-10-01T15:50:00.000Z","collectedAt":"2026-10-01T16:24:26.193Z","updatedAt":"2026-10-01T16:24:32.310Z","source":{"id":"botclaw-editorial","name":"BotClaw","origin":"editorial"},"summaryMethod":"editor","applicability":{"version":null,"platforms":[],"requirements":["核对当前账户访问、方案与任务条件。"]},"verification":{"kind":"official_documentation","checkedAt":"2026-10-01T15:50:00.000Z","urls":["https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse","https://docs.openclaw.ai/gateway/security","https://hermes-agent.nousresearch.com/docs/user-guide/security","https://help.openai.com/en/articles/20001530-getting-started-with-your-dot"],"limitations":["有来源的事实与本站编辑判断；未运行产品账户、推理或 agent 性能测试。"]},"body":{"introduction":"从委托任务出发。读取草稿、发送给接收方和支付账单后果不同，需要各自的规则。","sections":[{"heading":"写成可执行的边界","paragraphs":["“对外邮件先询问”需要定义外部范围、允许接收方、附件是否包含以及批准何时失效。无人值守前先明确这些条件。"]},{"heading":"检查在哪里执行","paragraphs":["区分自然语言指导、工具权限和操作系统边界。使用测试数据核对一个允许动作和一个应阻止动作，保存活动记录与具体批准内容。"]},{"heading":"最小起始规则","paragraphs":[],"steps":["允许读取指定测试工作区。","新增接收方、购买或难恢复删除需要明确批准。","后台任务设置停止条件和重试上限。"]}]},"editorial":{"section":"topic","productIds":["muse","openclaw","hermes","dot","cue"],"relatedIds":["wf-minimum-permissions","obs-topic-isolation"],"sourceDate":null,"version":null},"indexable":true,"supersededBy":null,"review":{"state":"current","pendingChanges":0,"policyDays":7}}}