Menu
部署后端到生产 (deploy.sh -b)。当用户说「部署后端」「上线」「deploy」「发后端」「重启后端」「跑迁移上线」时使用。含「合并≠上线」铁律、managed 迁移、健康度自动回滚、ssh 复验。
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
管理首页行动卡片 — 创建、查询、完成、归档。当用户想把某条建议固化到首页、查看当前行动计划、标记完成，使用这个 skill。
Assess chronic disease risk — cardiovascular (modified Framingham), metabolic syndrome (IDF criteria), and 90-day trend analysis. Integrates wearable data (Garmin), labs, vitals, and genetic markers for personalized risk scoring with confidence intervals.
Family health management - medical reports, medications, review calendar, family daily health check. Use when user asks about family members' health, medical exams, medications, checkup schedules, or wants to manage health for parents/spouse/children.
Query and analyze user genetic test data (基因检测). Cross-reference genetics with medical reports, Garmin wearable data, and current health status for personalized nutrition, exercise, drug sensitivity, disease risk, and sleep advice.
Get AI health analysis, daily recommendations, health trend predictions, and health scores. Use when the user asks for health advice, trend analysis, risk assessment, or wellness recommendations.
综合查询健康数据 — Digital Health Twin 快照、Safety 告警、长期趋势、Orchestrator 多专家分析。当用户想了解自己的整体健康状况、趋势变化、安全风险时使用。
| name | backend-deploy |
| description | 部署后端到生产 (deploy.sh -b)。当用户说「部署后端」「上线」「deploy」「发后端」「重启后端」「跑迁移上线」时使用。含「合并≠上线」铁律、managed 迁移、健康度自动回滚、ssh 复验。 |
后端上线唯一入口:./deploy.sh -b。异步执行,触发后切别的活,别同步等。
把代码合进 main 不会让它生效——生产进程还是旧的。必须 deploy.sh -b 才会 pull + 重启。
(本会话踩过:agenda 路由合并了没部署 → 手机「加载失败」,生产 /agenda/today 返回 404 而非 401。)
cd /Users/liqiuhua/work/personal/health-llm-driven
git push origin main # 先 push(deploy 在服务器 git pull)
./deploy.sh -b # 后台跑
deploy.sh -b 自动做:git pull → 应用 managed 迁移(backend/migrations/managed/*)→ 重启 health-backend + Celery worker+beat → DB 备份 → 同步 backend/skills/*/SKILL.md 到 OpenClaw 网关 → 跑 system_health_score.py。
健康度门:阈值 DEPLOY_SCORE_THRESHOLD=35(满分 60,skip-tests)。低于阈值自动回滚到上一版本。正常输出 健康度: 60/60 ✅ PASS + Skills 同步: 本地 N = 线上 N。
# 路由活着 = 401(需鉴权);404 = 进程是旧的(没重启成功)
ssh root@39.98.206.178 'curl -s -o /dev/null -w "%{http_code}\n" http://127.0.0.1:8000/api/v1/<新路由>'
401 = 上线成功;404 = 排查(进程没重启 / 路由没注册)。
git pull 失败。修:服务器上 git checkout -B main origin/main。push_code 对任何 dirty/untracked 树会硬退出。cd mobile && npm run generate-types 重生成移动端类型(否则手写类型静默漂移)。safety-gate(派 safety-privacy-reviewer,GO 才部署)。doc-drift-fix(CI 会卡 doc-drift)。.env 同步坑(deploy_backend 调 sync_env):-b 会把根 .env($SCRIPT_DIR/.env,非 backend/.env)scp 到 prod:backend/.env(先备份远端 .env,留 20 份)。所以必须从有真 prod 根 .env 的目录跑(根 .env 形态=含 DEPLOY_SERVER/DEVICE_ENCRYPTION_KEY/DATABASE_URL=postgresql)。从干净 worktree 跑要先把根 .env 拷进 worktree,否则 sync_env 推空/dev .env 污染 prod(backend/.env 是 dev/sqlite,别混)。并发 agent 翻分支时,worktree 跑更稳(git push 在 main 上是空操作)。ENABLE/FORCE ROW LEVEL SECURITY。生产 DB 连接角色必须是非 superuser(superuser 绕过 RLS → 隔离退化为仅应用层)。部署后看 journalctl -u health-backend 有无 [SECURITY] ... superuser ... RLS 被绕过 告警;有则把 prod DATABASE_URL 的角色换成非 superuser(如 health_user)。39.98.206.178(SSH 22)· /opt/health-app/ · systemd health-backend → health-api.executor.life · PostgreSQL 同机ssh root@39.98.206.178 "journalctl -u health-backend -n 50 --no-pager"-f 仅前端 · -a 全部 · -e 同步 .env 重启 · -r 仅重启 · -s 看状态 · -l 看日志前端/移动端不走这个:Web 用
deploy.sh -f;移动端 JS 走mobile-ota,原生/发版走mobile-testflight-release。