Deploy Checklist: [Service/Release]【免费下载链接】knowledge-work-pluginsOpen source repository of plugins primarily intended for knowledge workers to use in Claude Cowork项目地址: https://gitcode.com/GitHub_Trending/kn/knowledge-work-pluginsDate:[Date] |Deployer:[Name]Pre-DeployAll tests passing in CICode reviewed and approvedNo known critical bugs in releaseDatabase migrations tested (if applicable)Feature flags configured (if applicable)Rollback plan documentedOn-call team notifiedDeployDeploy to staging and verifyRun smoke testsDeploy to production (canary if available)Monitor error rates and latency for 15 minVerify key user flowsPost-DeployConfirm metrics are nominalUpdate release notes / changelogNotify stakeholdersClose related ticketsRollback TriggersError rate exceeds [X]%P50 latency exceeds [X]ms[Critical user flow] fails清单头部固定携带 Service/Release 名称、日期与部署人Deployer便于后续审计与复盘。四个区块的职责分工如下 ### Pre-Deploy上线前的质量与审批闸门 这一阶段回答「现在是否可以放行」 - **All tests passing in CI**以 CI 全绿作为最基本的准入条件避免把已知失败带上生产 - **Code reviewed and approved**确认变更已经过代码评审并获得批准 - **No known critical bugs in release**明确本版本没有已知的严重缺陷 - **Database migrations tested (if applicable)**条件项——只有本次变更包含数据库迁移时才需要勾选。数据库迁移是回滚最困难的变更类型必须先在 staging 实测 - **Feature flags configured (if applicable)**条件项——使用功能开关的变更需确认开关默认状态、灰度分组与关闭开关的兜底行为均已配置 - **Rollback plan documented**把回滚方案以文档形式固化为部署的一部分 - **On-call team notified**上线期间值班团队知情一旦出现异常可立即响应。 ### Deploy从 staging 到生产的渐进式验证 这一阶段描述「如何稳妥地把变更推上线」 - **Deploy to staging and verify**先在 staging 环境部署并验证确认行为符合预期 - **Run smoke tests**执行冒烟测试快速验证核心链路可用。该检查项与工程插件另一技能 testing-strategy[engineering/skills/testing-strategy/SKILL.md](https://link.gitcode.com/i/42bf759202ac2f7b8b29faaaf50d31e5)中针对基础设施场景推荐的 smoke tests 一脉相承 - **Deploy to production (canary if available)**生产部署优先采用金丝雀canary方式先放量部分流量观察再全量发布 - **Monitor error rates and latency for 15 min**部署后至少观察 15 分钟的报错率与延迟覆盖新版本刚上线的风险窗口 - **Verify key user flows**主动验证关键用户路径登录、下单、读写等而不是只依赖被动监控。 ### Post-Deploy收尾与闭环 - **Confirm metrics are nominal**确认核心指标回到/保持正常基线 - **Update release notes / changelog**同步更新发布说明与变更日志 - **Notify stakeholders**向相关方通报上线结果 - **Close related tickets**关闭与之关联的任务单形成完整闭环。 ### Rollback Triggers部署前就写好的回滚红线 回滚触发条件被刻意放在**部署之前**确定而非事故发生时临时讨论 - Error rate exceeds [X]%报错率超过预设阈值如 1% - P50 latency exceeds [X]msP50 延迟超过预设毫秒数如 500ms - [Critical user flow] fails关键用户流程不可用。 其中 [X] 与 [Critical user flow] 是需要团队根据自身服务基线与业务关键路径预先填写的占位值。接入监控连接器后Agent 可以直接从当前监控基线自动预填这些阈值见下文「连接器增强」。 ## 按需定制三类常见场景的清单演化 deploy-checklist 的定制逻辑在 [SKILL.md](https://link.gitcode.com/i/b57a26cd4069643ef4a2485731ea08c5) 的 Customization 一节中有明确说明——向 Agent 描述你的部署特征清单会自动追加对应检查项 | 场景描述 | 清单演化 | |---|---| | We use feature flags | 追加功能开关验证步骤确认开关默认值、灰度比例、kill-switch 关闭路径、相关日志与指标埋点 | | This includes a database migration | 追加迁移专项检查迁移脚本在 staging 的完整演练、数据校验、迁移的可逆性/回滚脚本、锁表与停机窗口评估 | | This is a breaking API change | 追加下游消费者通知步骤识别 API 调用方、发布变更公告、确认消费者升级窗口与兼容策略 | 值得注意的是**定制化不止于单次对话**。工程插件支持通过本地设置文件固化团队默认部署流程[engineering/README.md](https://link.gitcode.com/i/3e6cd315402301137f46cb3457cc6d62) 的 Settings 一节给出的 engineering/.claude/settings.local.json 示例中包含 json { name: Your Name, title: Software Engineer, team: Your Team, company: Your Company, techStack: [Python, TypeScript, PostgreSQL, AWS], defaultBranch: main, deployProcess: canary }【免费下载链接】knowledge-work-pluginsOpen source repository of plugins primarily intended for knowledge workers to use in Claude Cowork项目地址: https://gitcode.com/GitHub_Trending/kn/knowledge-work-plugins创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考