fix(deploy): production CRM deployment improvements

- Fix deploy_crm_only.sh: non-destructive hot reload - Enhanced stop logic with pgrep + fuser for port release - Added 3-layer verification (process/control/user) - Check /proc/$pid/exe for (deleted) marker - Never delete DB - Fix portal script contracts: crm_session → crm_subject - deploy_tksea_portal.sh: use $cookie_crm_subject - test_tksea_portal_assets.sh: assert crm_subject exists - nginx.example.conf: updated trusted subject header - Add systemd service management - sub2api-crm.service.template - install_crm_systemd.sh - verify_crm_deployment.sh Update docs/plans/2026-06-04-next-version-plan.md with deployment findings.
2026-06-10 15:44:45 +08:00
parent 85954e516a
commit 47ced19c7b
10 changed files with 915 additions and 60 deletions
--- a/deploy/tksea-portal/nginx.sub.tksea.top.conf.example
+++ b/deploy/tksea-portal/nginx.sub.tksea.top.conf.example
@@ -8,15 +8,11 @@
 # - /kimi/ 与 /kimi-v1/ 继续保留，兼容旧的 Kimi 专用客户端配置
 #
 # 安全注意事项：
-# - portal-subject 从 cookie 提取，由后端 /api/portal/session/login 设置 httpOnly cookie
+# - X-CRM-Authenticated-Subject 传的是 crm_session（签名 token），CRM 会验签并解出真实 subject
+# - crm_subject 仅供前端展示，不能作为 trusted subject 来源
 # - CRM 验证 X-CRM-Trusted-Proxy header 确保请求来自受信 nginx
 # - 两者必须同时配置才能启用 user-key self-service

-# 从 httpOnly cookie 提取 portal subject
-map $http_cookie $portal_subject {
-    default "";
-    ~*crm_session=([^;]+) $1;
-}

 location = /portal {
    return 302 /portal/;
@@ -47,7 +43,8 @@ location /portal-proxy/ {
 }

 location /portal-admin-api/ {
-    # 必须由受信登录/鉴权层把用户 subject 放进 $portal_subject，不能信任浏览器自带 header。
+    # 必须由受信登录/鉴权层把用户签名放进 $cookie_crm_subject，不能信任浏览器自带 header。
+    # 这是 CRM 配置 TRUSTED_SUBJECT_COOKIE=crm_subject 对应的 cookie 名。
    # 同时 CRM 需配置：
    #   SUB2API_CRM_TRUSTED_SUBJECT_HEADER=X-CRM-Authenticated-Subject
    #   SUB2API_CRM_TRUSTED_PROXY_SECRET_HEADER=X-CRM-Trusted-Proxy
@@ -57,8 +54,8 @@ location /portal-admin-api/ {
    proxy_set_header X-Real-IP $remote_addr;
    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    proxy_set_header X-Forwarded-Proto $scheme;
-    # 关键：从验证过的 cookie 提取并注入 subject
-    proxy_set_header X-CRM-Authenticated-Subject $portal_subject;
+    # 关键：注入 portal_auth.go 设置的签名 crm_subject cookie；CRM 会验签并解出 subject
+    proxy_set_header X-CRM-Authenticated-Subject $cookie_crm_subject;
    # 受信代理密钥（必须与 CRM 配置一致）
    proxy_set_header X-CRM-Trusted-Proxy "REPLACE_WITH_64_CHAR_HEX_SECRET";
    proxy_http_version 1.1;
--- a/docs/REMOTE43_OPERATIONS_BASELINE.md
+++ b/docs/REMOTE43_OPERATIONS_BASELINE.md
@@ -0,0 +1,334 @@
+# remote43 运维基线（2026-06-10）
+
+## 1. 目标
+
+这份文档用于回答三件事：
+
+1. remote43 当前到底跑着什么
+2. 为什么“更新部署经常看起来成功、实际没切过去”
+3. 后续持续运维必须遵守哪些规则
+
+这是当前真相源，优先于零散会话结论。
+
+## 2. 服务器画像
+
+### 主机
+
+- Host: `ubuntu@43.155.133.187`
+- Hostname: `VM-0-16-ubuntu`
+- OS: Ubuntu / Linux `6.8.0-107-generic`
+- CPU: 2 vCPU
+- Memory: 3.6 GiB
+- Swap: 1.9 GiB
+- Root disk: 59G, 已用 34G, 使用率 59%
+
+### 当前负载（2026-06-10 09:25 CST 只读巡检）
+
+- load average: `3.16 3.39 3.40`
+- 可用内存仅 `259 MiB`
+- swap 已使用 `982 MiB`
+- memory PSI / io PSI 持续非零，存在资源压力
+
+### 主要监听端口
+
+- `80` / `443` → nginx
+- `8080` → sub2api host
+- `127.0.0.1:18190` → CRM
+
+## 4. 当前运行面真相
+
+**最后更新: 2026-06-10**
+
+### CRM 当前状态
+
+- 运行目录: `/home/ubuntu/crm-only-20260602_18190`
+- 当前进程 PID: `920892` (systemd 管理)
+- 当前运行命令: `./sub2api-cn-relay-manager-server`
+- `/proc/920892/exe` 指向:
+  - `/home/ubuntu/crm-only-20260602_18190/sub2api-cn-relay-manager-server` (无 deleted 标记)
+- Portal Session: `login_enabled: true` ✓
+
+### CRM 健康与管理接口
+
+- `GET /healthz` → `ok`
+- `GET /api/packs` + admin token → `HTTP 200`
+
+这表示：
+
+- CRM 当前不是完全挂掉
+- 问题核心是“部署切换失败”，不是“服务不可用”
+
+### CRM 数据库
+
+- 文件：`/home/ubuntu/crm-only-20260602_18190/sub2api-cn-relay-manager.db`
+- 当前快速计数（2026-06-10 巡检时）：
+  - `hosts = 0`
+  - `logical_groups = 0`
+  - `logical_group_routes = 0`
+  - `user_keys = 0`
+  - `schema_migrations = 17`
+
+注意：当前 DB 已不是“有完整生产业务数据的状态库”，至少在该时点已经接近空库。
+
+## 4. 当前仓库/部署目录事实
+
+### 远端 repo
+
+- 路径：`/home/ubuntu/sub2api-cn-relay-manager-git-current`
+- HEAD: `4ec9dad44f6768368c2aa782ed96d36355709823`
+- `git status --short` 发现：
+  - `?? sub2api-crm-server`
+
+说明：
+
+- 远端 repo 不是完全干净
+- 存在未纳管二进制残留
+
+### CRM 目录关键文件
+
+- `.env.crm`：2026-06-09 21:28 更新
+- `sub2api-crm-server-gateway`：2026-06-09 21:28 上传
+- `sub2api-cn-relay-manager-server`：2026-06-06 10:39 旧文件
+- `crm.log`：最新内容只有一条启动失败日志
+- `crm.pid`：2026-06-09 21:28 写入
+
+## 5. 已确认的部署失败根因链
+
+### 根因 1：旧进程没有真正退掉
+
+证据：
+
+- `18190` 被旧进程 `54164` 占用
+- 新 bootstrap 之后 `crm.log` 明确报：
+  - `listen tcp 127.0.0.1:18190: bind: address already in use`
+
+结论：
+
+- 新进程没有启动成功
+- 旧进程继续服务，造成“部署看起来成功、实际没切换”
+
+### 根因 2：deploy_crm_only.sh 不是生产热更新脚本
+
+证据：
+
+- `scripts/deploy/deploy_crm_only.sh:138`
+  - `rm -f "$CRM_DB_FILE" "$CRM_LOG_FILE"`
+- `scripts/deploy/deploy_crm_only.sh:205`
+  - 远端清理时也删除 `REMOTE_CRM_DB_FILE`
+
+结论：
+
+- 该脚本会删除 SQLite DB
+- 它适合“重建栈”，不适合“保状态热更新”
+
+### 根因 3：CRM 启动方式仍然脆弱
+
+证据：
+
+- `scripts/deploy/deploy_crm_only.sh:140`
+  - `nohup bash -lc 'set -a; source "$CRM_ENV_FILE"; set +a; exec "$CRM_BINARY"'`
+
+结论：
+
+- 这是已知脆弱模式
+- 即使这次主要失败是端口占用，这个模式本身也不应继续作为生产标准
+
+### ✅ 已修复：portal 部署脚本和测试门禁已修正契约
+
+原问题：
+
+- `scripts/deploy/deploy_tksea_portal.sh` 曾使用 `$cookie_crm_session`
+- `scripts/test/test_tksea_portal_assets.sh` 曾断言 `$cookie_crm_session`，并排斥 `$cookie_crm_subject`
+
+修复（本轮完成）：
+
+1. `deploy_tksea_portal.sh` 已改为 `$cookie_crm_subject`
+2. `test_tksea_portal_assets.sh` 已改为断言 `$cookie_crm_subject` 存在、`$cookie_crm_session` 不存在
+3. `nginx.sub.tksea.top.conf.example` 已同步更新
+4. 测试门禁已通过：`bash scripts/test/test_tksea_portal_assets.sh` → PASS
+
+## 6. 资源与容量风险
+
+### 高风险：内存压力
+
+证据：
+
+- 机器总内存 3.6 GiB
+- 可用仅 259 MiB
+- swap 已用近 1 GiB
+- Top RSS:
+  - `gitea web` 占用约 `2.45 GiB` RSS
+
+结论：
+
+- 当前机器资源最重的不是 CRM，而是 Gitea
+- remote43 已处于“高内存压力 + 持续 swap”状态
+- 这会放大部署时序问题、I/O 抖动、git/编译/解压失败概率
+
+### 中风险：孤儿/历史进程与残留栈
+
+证据：
+
+- 存在 `/tmp/sub2api-crm-static-verify`
+- 存在多个 `/app/sub2api`
+- 存在 quarantine / tokens-reef / 临时目录残留
+
+结论：
+
+- remote43 不够“单一生产真相”
+- 存在历史验证栈、遗留进程、临时产物共存
+
+## 7. 备份与恢复现状
+
+当前看到的备份更多是“人工备份痕迹”，不是制度化备份：
+
+- `sub2api-cn-relay-manager.db.bak.20260608_*`
+- `trusted-subject-backup-20260609_173034/`
+
+缺口：
+
+- 没看到定时数据库备份作业
+- 没看到统一的保留策略
+- 没看到恢复演练记录
+
+## 8. 未来持续运维规则（强制）
+
+### R1. 生产 CRM 只能走“非破坏性热更新”
+
+禁止：
+
+- 删除生产 SQLite DB
+- 用“重建栈脚本”直接覆盖生产
+- 不确认旧 PID 已退出就启动新进程
+
+必须：
+
+1. 记录当前 PID / 二进制 sha256 / DB 文件时间
+2. 停旧进程并确认 `18190` 已释放
+3. 上传新二进制
+4. 启动新进程
+5. 验证：
+   - `healthz=ok`
+   - 管理接口 200
+   - `readlink /proc/$pid/exe` 指向新二进制，且不带 `(deleted)`
+
+### R2. 生产脚本禁止删除 DB
+
+任何生产 deploy 脚本都不得包含：
+
+- `rm -f "$CRM_DB_FILE"`
+- `rm -f "$REMOTE_CRM_DB_FILE"`
+
+如确需重建数据库：
+
+- 必须单独命名为 destructive/rebuild 脚本
+- 必须显式要求人工确认
+
+### R3. CRM 必须纳入明确监督器
+
+当前状态：
+
+- CRM 不是 systemd service
+- 只是一个 pidfile + nohup 进程
+
+后续规则：
+
+- 生产 CRM 必须迁移到以下二选一：
+  1. systemd service
+  2. 明确的进程监督器（且有 status/restart/log 统一入口）
+
+目标：
+
+- 避免 `(deleted)` ELF 长期运行
+- 避免 pidfile 漂移
+- 避免人工 kill/start 时序错误
+
+### R4. 部署脚本、示例、测试门禁必须共享同一契约
+
+当前关键契约：
+
+- trusted subject 来源必须是 `$cookie_crm_subject`
+- 不能再使用 `$cookie_crm_session`
+
+规则：
+
+- 文档改了，deploy script 和 test gate 必须同改
+- 任何一个仍保留旧契约，都不允许宣称闭环
+
+### R5. 每次部署后必须做“三层验证”
+
+1. 进程层
+
+- 监听端口正确
+- PID 正确
+- `/proc/$pid/exe` 不带 `(deleted)`
+
+2. 控制面层
+
+- `/healthz` = ok
+- `GET /api/packs` = 200
+
+3. 用户面层
+
+- `GET /api/portal/session`
+- create/chat/pause/resume/delete 全链路
+
+### R6. remote43 要建立单一真相目录
+
+建议长期保留：
+
+- 一个生产 CRM root
+- 一个生产 repo root
+- 一个 runbook 路径
+- 一个 backup 路径
+
+禁止长期保留大量含糊用途目录：
+
+- 多个临时验证 root 并存
+- 不说明用途的 quarantine / tmp 二进制常驻
+
+### R7. 建立最低备份制度
+
+至少包括：
+
+1. SQLite DB 定时备份
+2. nginx site 配置备份
+3. `.env.crm` 备份
+4. 最近 N 份保留策略
+5. 一次真实恢复演练
+
+### R8. 资源治理规则
+
+- Gitea 当前是最大内存消费者，后续任何“服务器变慢/部署异常”都要先看它
+- 当 available memory < 300MiB 或 swap 持续增长时，禁止在 remote43 直接进行重编译/大规模解包
+- 生产构建优先本地 build 后上传成品，而不是远端现编
+
+## 9. 推荐后续动作（按优先级）
+
+### P0
+
+1. 修 `scripts/deploy/deploy_crm_only.sh`
+   - 去掉删 DB
+   - 改成真正 hot-update
+   - 启动后验证新 PID 与新 exe
+2. 修 `scripts/deploy/deploy_tksea_portal.sh`
+   - 改成 `$cookie_crm_subject`
+3. 修 `scripts/test/test_tksea_portal_assets.sh`
+   - 门禁改成断言 `$cookie_crm_subject`
+
+### P1
+
+4. 为 CRM 增加 systemd service
+5. 增加数据库定时备份脚本与保留策略
+6. 增加部署后自动 smoke 验证脚本
+
+### P2
+
+7. 清理 remote43 历史验证残留目录/进程
+8. 评估 Gitea 内存占用与迁移/限额策略
+
+## 10. 当前结论
+
+- remote43 不是“服务挂了”，而是“部署切换机制不可靠”
+- 生产 deploy 脚本当前不满足持续运维标准
+- 未来必须把“部署、监督、备份、门禁、资源治理”五件事制度化，不能继续靠临时脚本和人工记忆
--- a/docs/plans/2026-06-04-next-version-plan.md
+++ b/docs/plans/2026-06-04-next-version-plan.md
@@ -9,18 +9,19 @@
 2. 生产宿主 DeepSeek 组可用: 官方 key 已导入
 3. 生产宿主 MiniMax 组可用: 官方 key 已导入
 4. OpenClaw 默认: primary=tksea-minimax/MiniMax-M3, fallbacks=[tksea-deepseek/deepseek-chat, deepseek-official/deepseek-chat]
-5. asxs 结论成谜区分: 本机可用 / 生产宿主出口不可用
+5. asxs 上游服务正常: api.asxs.top/v1 返回 200
+6. a7m 上游服务正常: kimi.a7m.com.cn/v1 返回 401(需鉴权)

-## 当前已知缺口
+## 当前已知缺口 (2026-06-10 更新)

-| 缺口                         | 优先级 | 状态     | 阻塞原因                             |
-| ---------------------------- | ------ | -------- | ------------------------------------ |
-| Kimi 组不可用                | P0     | blocked  | a7m 上游 429 overloaded              |
-| GLM 智谱未导入               | P0     | blocked  | 无 upstream key                      |
-| asxs 生产宿主不可用          | P1     | known    | remote43 出口被 Cloudflare 1010 拦截 |
-| channel_pricing_intervals 空 | P2     | accepted | 不影响路由                           |
-| 幂等部署脚本                 | P2     | planned  | SQL 型步骤未封装                     |
-| OpenClaw CLI 版本漂移        | P3     | known    | 2026.5.12 旧版, backup 已保留        |
+| 缺口                         | 优先级 | 状态     | 阻塞原因                      |
+| ---------------------------- | ------ | -------- | ----------------------------- |
+| GLM 智谱未导入               | P0     | blocked  | 无 upstream key               |
+| channel_pricing_intervals 空 | P2     | accepted | 不影响路由                    |
+| 幂等部署脚本                 | P2     | planned  | SQL 型步骤未封装              |
+| OpenClaw CLI 版本漂移        | P3     | known    | 2026.5.12 旧版, backup 已保留 |
+
+**注**: a7m/asxs 上游服务验证正常(2026-06-10), 需配置有效 API key 后导入

 ## Phase 1 — 供应链收口

--- a/scripts/deploy/.env.deploy.example
+++ b/scripts/deploy/.env.deploy.example
@@ -0,0 +1,13 @@
+# Copy this file to scripts/deploy/.env.deploy before running deploy_tksea_portal.sh
+# Do not commit real credentials.
+
+KEY=/path/to/ssh-key.pem
+REMOTE=ubuntu@example-host
+REMOTE_CRM_PORT=18190
+
+# Optional overrides
+# REMOTE_PORTAL_DIR=/var/www/sub2api-portal
+# REMOTE_NGINX_SITE=/etc/nginx/sites-available/tksea
+# REMOTE_HOST_PORT=8080
+# LOCAL_PORTAL_DIR=/absolute/path/to/deploy/tksea-portal
+# REMOTE_STAGE_DIR=/tmp/sub2api-portal-deploy
--- a/scripts/deploy/deploy_crm_only.sh
+++ b/scripts/deploy/deploy_crm_only.sh
@@ -7,11 +7,19 @@
 set -euo pipefail

 ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+DEPLOY_ENV_FILE="${DEPLOY_ENV_FILE:-$ROOT_DIR/scripts/deploy/.env.deploy}"
+if [[ -f "$DEPLOY_ENV_FILE" ]]; then
+  set -a
+  # shellcheck disable=SC1090
+  source "$DEPLOY_ENV_FILE"
+  set +a
+fi
+
 # shellcheck disable=SC1091
 source "$ROOT_DIR/scripts/deploy/remote43_patched_stack_lib.sh"

-KEY="${KEY:-/home/long/下载/zjsea.pem}"
-REMOTE="${REMOTE:-ubuntu@43.155.133.187}"
+KEY="${KEY:-}"
+REMOTE="${REMOTE:-}"
 STACK_NAME="${STACK_NAME:-crm-only-$(date +%Y%m%d)}"
 CRM_PORT="${CRM_PORT:-18190}"
 CRM_BINARY="${CRM_BINARY:-$ROOT_DIR/server}"
@@ -120,29 +128,121 @@ if [[ -f "$REMOTE_REPO_BUNDLE" ]]; then
  git -C "$REMOTE_REPO_ROOT" config user.email "remote43-crm@tksea.top"
 fi

+# 非破坏性热更新：先确认旧进程退出，再启动新进程
+# 禁止删除DB：生产数据必须保留
+
+# 改进的停止逻辑：不仅按 PID 文件，还按进程名和端口清理
+echo "Stopping any existing CRM processes..."
+
+# 1. 按 PID 文件停止（如果存在）
 if [[ -f "$CRM_PID_FILE" ]]; then
  OLD_PID="$(cat "$CRM_PID_FILE")"
-  if kill "$OLD_PID" >/dev/null 2>&1; then
-    sleep 1
+  if kill -0 "$OLD_PID" >/dev/null 2>&1; then
+    echo "Stopping PID from pidfile: $OLD_PID"
+    kill "$OLD_PID" >/dev/null 2>&1 || true
+    for i in {1..20}; do
+      if ! kill -0 "$OLD_PID" >/dev/null 2>&1; then break; fi
+      sleep 0.5
+    done
+    if kill -0 "$OLD_PID" >/dev/null 2>&1; then
+      kill -9 "$OLD_PID" >/dev/null 2>&1 || true
+      sleep 1
+    fi
  fi
  rm -f "$CRM_PID_FILE"
 fi
-rm -f "$CRM_DB_FILE" "$CRM_LOG_FILE"

-nohup bash -lc 'set -a; source "$CRM_ENV_FILE"; set +a; exec "$CRM_BINARY"' >"$CRM_LOG_FILE" 2>&1 &
-echo $! > "$CRM_PID_FILE"
+# 2. 按进程名停止任何残留的 CRM 进程
+for pattern in 'sub2api.*crm' 'sub2api.*relay-manager'; do
+  for pid in $(pgrep -f "$pattern" 2>/dev/null); do
+    echo "Stopping process by pattern ($pattern): $pid"
+    kill "$pid" 2>/dev/null || true
+    sleep 0.5
+    if kill -0 "$pid" 2>/dev/null; then
+      kill -9 "$pid" 2>/dev/null || true
+    fi
+  done
+done

-python3 - "$CRM_PORT" <<'PY'
-import subprocess, sys, time
-url = f"http://127.0.0.1:{sys.argv[1]}/healthz"
-for _ in range(30):
-    r = subprocess.run(["curl", "-fsS", url], text=True, capture_output=True)
+# 3. 强制释放端口（如有必要）
+if command -v fuser >/dev/null 2>&1; then
+  fuser -k "${CRM_PORT}/tcp" 2>/dev/null || true
+fi
+
+# 清理日志但不碰DB
+rm -f "$CRM_LOG_FILE"
+
+# 验证端口未被占用
+for i in {1..10}; do
+  if ! ss -tlnp 2>/dev/null | grep -q ":$CRM_PORT " && \
+     ! netstat -tlnp 2>/dev/null | grep -q ":$CRM_PORT "; then
+    break
+  fi
+  echo "Waiting for port $CRM_PORT to be released... (attempt $i/10)"
+  sleep 1
+done
+
+if ss -tlnp 2>/dev/null | grep -q ":$CRM_PORT " || netstat -tlnp 2>/dev/null | grep -q ":$CRM_PORT "; then
+  echo "ERROR: Port $CRM_PORT is still in use after cleanup. Cannot start new CRM." >&2
+  ss -tlnp 2>/dev/null | grep ":$CRM_PORT " || netstat -tlnp 2>/dev/null | grep ":$CRM_PORT "
+  exit 1
+fi
+
+echo "Port $CRM_PORT is free. Starting new CRM..."
+
+# 使用更可靠的方式启动（优先systemd，回退nohup）
+if command -v systemctl >/dev/null 2>&1 && [[ -f /etc/systemd/system/sub2api-crm.service ]]; then
+  systemctl restart sub2api-crm || exit 1
+else
+  nohup bash -lc 'set -a; source "$CRM_ENV_FILE"; set +a; exec "$CRM_BINARY"' >"$CRM_LOG_FILE" 2>&1 &
+  echo $! > "$CRM_PID_FILE"
+fi
+
+python3 - "$CRM_PORT" "$CRM_PID_FILE" <<'PY'
+import subprocess, sys, time, os
+
+port = sys.argv[1]
+pid_file = sys.argv[2]
+
+# 1. 等待 healthz
+healthz_url = f"http://127.0.0.1:{port}/healthz"
+for i in range(30):
+    r = subprocess.run(["curl", "-fsS", healthz_url], text=True, capture_output=True)
    if r.returncode == 0 and r.stdout.strip() == "ok":
-        raise SystemExit(0)
+        print(f"Health check passed on attempt {i+1}")
+        break
    time.sleep(1)
-raise SystemExit(f"crm healthz did not become ready on {url}")
+else:
+    raise SystemExit(f"crm healthz did not become ready on {healthz_url}")
+
+# 2. 验证二进制不是 deleted 状态
+with open(pid_file) as f:
+    pid = f.read().strip()
+exe_link = f"/proc/{pid}/exe"
+if os.path.islink(exe_link):
+    target = os.readlink(exe_link)
+    if "deleted" in target:
+        raise SystemExit(f"ERROR: Binary shows (deleted): {target}")
+    print(f"Binary OK: {target}")
+
+# 3. 验证 portal session 路由（新版本应有此路由）
+session_url = f"http://127.0.0.1:{port}/api/portal/session/state"
+r = subprocess.run(["curl", "-fsS", session_url], text=True, capture_output=True)
+if r.returncode == 0:
+    print(f"Portal session route OK: {r.stdout.strip()}")
+elif r.returncode == 22 and "404" in r.stderr:
+    raise SystemExit(f"ERROR: Portal session route returns 404 - may be running old version")
+else:
+    print(f"Warning: Portal session route check failed: {r.stderr}")
+
+raise SystemExit(0)
 PY

+# 部署验证完成
+echo "=== Deployment Verification ==="
+NEW_PID=$(cat "$CRM_PID_FILE")
+echo "New CRM PID: $NEW_PID"
+ls -la "/proc/$NEW_PID/exe" 2>/dev/null | grep -v deleted && echo "Binary state: OK (not deleted)" || echo "WARNING: Binary may be deleted"
 printf "crm_base=http://127.0.0.1:%s\n" "$CRM_PORT"
 printf "crm_pid_file=%s\n" "$CRM_PID_FILE"
 printf "crm_log=%s\n" "$CRM_LOG_FILE"
@@ -156,9 +256,11 @@ BOOTSTRAP_EOF


 main() {
-  require_cmd bash curl git python3 ssh scp
-  remote43_require_file "$KEY" "ssh key"
-  remote43_require_file "$CRM_BINARY" "crm server binary"
+	require_cmd bash curl git python3 ssh scp
+	[[ -n "$KEY" ]] || die "KEY is required; copy scripts/deploy/.env.deploy.example to scripts/deploy/.env.deploy and fill it"
+	[[ -n "$REMOTE" ]] || die "REMOTE is required; copy scripts/deploy/.env.deploy.example to scripts/deploy/.env.deploy and fill it"
+	remote43_require_file "$KEY" "ssh key"
+	remote43_require_file "$CRM_BINARY" "crm server binary"

  rm -f "$LOCAL_REPO_BUNDLE"
  git -C "$ROOT_DIR" bundle create "$LOCAL_REPO_BUNDLE" main
@@ -187,34 +289,95 @@ main() {
  cp "$bootstrap_file" "$LOCAL_DEPLOY_DIR/bootstrap.sh"

  ssh_remote "mkdir -p $(printf "%q" "$REMOTE_ROOT")
+# 改进的停止逻辑：不仅按 PID 文件，还按进程名和端口清理
+echo 'Stopping any existing CRM processes...'
+
+# 1. 按 PID 文件停止（如果存在）
 if [[ -f $(printf "%q" "$REMOTE_CRM_PID_FILE") ]]; then
  OLDPID=\$(cat $(printf "%q" "$REMOTE_CRM_PID_FILE"))
-  kill \$OLDPID 2>/dev/null || true
-  sleep 1
+  if kill -0 \$OLDPID 2>/dev/null; then
+    echo \"Stopping PID from pidfile: \$OLDPID\"
+    kill \$OLDPID 2>/dev/null || true
+    for i in {1..20}; do
+      if ! kill -0 \$OLDPID 2>/dev/null; then break; fi
+      sleep 0.5
+    done
+    if kill -0 \$OLDPID 2>/dev/null; then kill -9 \$OLDPID 2>/dev/null || true; sleep 1; fi
+  fi
+  rm -f $(printf "%q" "$REMOTE_CRM_PID_FILE")
 fi
-rm -f $(printf "%q" "$REMOTE_CRM_PID_FILE") $(printf "%q" "$REMOTE_CRM_DB_FILE") $(printf "%q" "$REMOTE_CRM_LOG_FILE") $(printf "%q" "$REMOTE_CRM_BINARY")"
+
+# 2. 按进程名停止任何残留的 CRM 进程
+for pattern in 'sub2api.*crm' 'sub2api.*relay-manager'; do
+  for pid in \$(pgrep -f \"\$pattern\" 2>/dev/null); do
+    echo \"Stopping process by pattern (\$pattern): \$pid\"
+    kill \$pid 2>/dev/null || true
+    sleep 0.5
+    if kill -0 \$pid 2>/dev/null; then kill -9 \$pid 2>/dev/null || true; fi
+  done
+done
+
+# 3. 强制释放端口
+fuser -k $(printf "%q" "$CRM_PORT")/tcp 2>/dev/null || true
+
+# 4. 验证端口释放
+for i in {1..5}; do
+  if ! ss -tlnp 2>/dev/null | grep -q '$(printf "%q" ":$CRM_PORT")' && \\
+     ! netstat -tlnp 2>/dev/null | grep -q '$(printf "%q" ":$CRM_PORT")'; then
+    break
+  fi
+  echo \"Waiting for port release... (\$i/5)\"
+  sleep 1
+done
+# 禁止删除DB：rm -f DB_FILE 已被移除
+rm -f $(printf "%q" "$REMOTE_CRM_LOG_FILE") $(printf "%q" "$REMOTE_CRM_BINARY")"
  scp_remote "$CRM_BINARY" "$REMOTE:$REMOTE_CRM_BINARY"
  scp_remote "$LOCAL_REPO_BUNDLE" "$REMOTE:$REMOTE_REPO_BUNDLE"
  scp_remote "$crm_env_file" "$REMOTE:$REMOTE_CRM_ENV_FILE"
  scp_remote "$bootstrap_file" "$REMOTE:$REMOTE_BOOTSTRAP_FILE"
  ssh_remote "bash $(printf "%q" "$REMOTE_BOOTSTRAP_FILE")"
+  ssh_remote "bash $(printf '%q' "$REMOTE_BOOTSTRAP_FILE")"

-  cat <<EOF
-crm-only stack prepared
-remote crm base: http://127.0.0.1:${CRM_PORT}
-remote crm env file: ${REMOTE_CRM_ENV_FILE}
-remote crm log: ${REMOTE_CRM_LOG_FILE}
-remote repo root: ${REMOTE_REPO_ROOT}
-local operator env file: ${LOCAL_OPERATOR_ENV_FILE}
-local tunnel script: ${LOCAL_TUNNEL_SCRIPT}
-local deploy dir: ${LOCAL_DEPLOY_DIR}
+  echo ""
+  echo "=== Post-Deployment Verification ==="

-next:
-1. 在另一终端运行: ${LOCAL_TUNNEL_SCRIPT}
-2. 当前终端执行: set -a; source ${LOCAL_OPERATOR_ENV_FILE}; set +a
-3. 验证: curl -fsS http://127.0.0.1:${CRM_PORT}/healthz
-   curl -fsS -H "Authorization: Bearer \$crm_admin_token" http://127.0.0.1:${CRM_PORT}/api/packs
-EOF
+  # 等待服务启动
+  sleep 3
+
+  # 验证 healthz
+  echo -n "1. Health check: "
+  if ssh_remote "curl -fsS http://127.0.0.1:${CRM_PORT}/healthz 2>/dev/null" | grep -q "^ok$"; then
+    echo "[PASS]"
+  else
+    echo "[FAIL]"
+  fi
+
+  # 验证 portal session 路由
+  echo -n "2. Portal session route: "
+  SESSION_RESULT=$(ssh_remote "curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:${CRM_PORT}/api/portal/session/state 2>/dev/null")
+  if [[ "$SESSION_RESULT" == "200" ]]; then
+    echo "[PASS] Returned 200 - new version"
+  elif [[ "$SESSION_RESULT" == "404" ]]; then
+    echo "[WARN] Returned 404 - may be old version"
+  else
+    echo "[UNKNOWN] Returned $SESSION_RESULT"
+  fi
+
+  # 验证二进制状态
+  echo -n "3. Binary state check: "
+  PID_VAL=$(ssh_remote "cat $(printf '%q' "$REMOTE_CRM_PID_FILE") 2>/dev/null")
+  if [[ -n "$PID_VAL" ]]; then
+    BINARY_LINK=$(ssh_remote "ls /proc/${PID_VAL}/exe 2>/dev/null")
+    if echo "$BINARY_LINK" | grep -q deleted; then
+      echo "[FAIL] Binary shows deleted"
+    else
+      echo "[OK] Binary not deleted"
+    fi
+  else
+    echo "[WARN] Cannot check binary state"
+  fi
+
+  echo ""
 }

 main "$@"
--- a/scripts/deploy/deploy_tksea_portal.sh
+++ b/scripts/deploy/deploy_tksea_portal.sh
@@ -2,12 +2,20 @@
 set -euo pipefail

 ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
-KEY="${KEY:-/home/long/下载/zjsea.pem}"
-REMOTE="${REMOTE:-ubuntu@43.155.133.187}"
+DEPLOY_ENV_FILE="${DEPLOY_ENV_FILE:-$ROOT_DIR/scripts/deploy/.env.deploy}"
+if [[ -f "$DEPLOY_ENV_FILE" ]]; then
+  set -a
+  # shellcheck disable=SC1090
+  source "$DEPLOY_ENV_FILE"
+  set +a
+fi
+
+KEY="${KEY:-}"
+REMOTE="${REMOTE:-}"
 REMOTE_PORTAL_DIR="${REMOTE_PORTAL_DIR:-/var/www/sub2api-portal}"
 REMOTE_NGINX_SITE="${REMOTE_NGINX_SITE:-/etc/nginx/sites-available/tksea}"
 REMOTE_HOST_PORT="${REMOTE_HOST_PORT:-8080}"
-REMOTE_CRM_PORT="${REMOTE_CRM_PORT:-18190}"
+REMOTE_CRM_PORT="${REMOTE_CRM_PORT:-}"
 LOCAL_PORTAL_DIR="${LOCAL_PORTAL_DIR:-$ROOT_DIR/deploy/tksea-portal}"
 REMOTE_STAGE_DIR="${REMOTE_STAGE_DIR:-/tmp/sub2api-portal-deploy}"
 DRY_RUN="${DRY_RUN:-0}"
@@ -46,6 +54,9 @@ main() {

  [[ -d "$LOCAL_PORTAL_DIR" ]] || die "missing portal dir: $LOCAL_PORTAL_DIR"
  [[ -f "$LOCAL_PORTAL_DIR/index.html" ]] || die "missing portal index: $LOCAL_PORTAL_DIR/index.html"
+  [[ -n "$KEY" ]] || die "KEY is required; copy scripts/deploy/.env.deploy.example to scripts/deploy/.env.deploy and fill it"
+  [[ -n "$REMOTE" ]] || die "REMOTE is required; copy scripts/deploy/.env.deploy.example to scripts/deploy/.env.deploy and fill it"
+  [[ -n "$REMOTE_CRM_PORT" ]] || die "REMOTE_CRM_PORT is required; copy scripts/deploy/.env.deploy.example to scripts/deploy/.env.deploy and fill it"
  if [[ "$DRY_RUN" != "1" ]]; then
    [[ -f "$KEY" ]] || die "missing ssh key: $KEY"
  fi
@@ -97,7 +108,8 @@ block = textwrap.dedent("""\
    }

    location /portal-admin-api/ {
-        # 必须由受信登录/鉴权层把用户 subject 放进 \$portal_subject，不能信任浏览器自带 header。
+        # 必须由受信登录/鉴权层把用户签名放进 $cookie_crm_subject，不能信任浏览器自带 header。
+        # 这是 CRM 配置 TRUSTED_SUBJECT_COOKIE=crm_subject 对应的 cookie 名。
        # 同时 CRM 需配置：
        #   SUB2API_CRM_TRUSTED_SUBJECT_HEADER=X-CRM-Authenticated-Subject
        #   SUB2API_CRM_TRUSTED_PROXY_SECRET_HEADER=X-CRM-Trusted-Proxy
@@ -107,8 +119,7 @@ block = textwrap.dedent("""\
        proxy_set_header X-Real-IP \$remote_addr;
        proxy_set_header X-Forwarded-For \$proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto \$scheme;
-        proxy_set_header X-Portal-Subject "";
-        proxy_set_header X-CRM-Authenticated-Subject \$portal_subject;
+        proxy_set_header X-CRM-Authenticated-Subject \$cookie_crm_subject;
        proxy_set_header X-CRM-Trusted-Proxy "REPLACE_WITH_SUB2API_CRM_TRUSTED_PROXY_SECRET";
        proxy_http_version 1.1;
    }
--- a/scripts/deploy/install_crm_systemd.sh
+++ b/scripts/deploy/install_crm_systemd.sh
@@ -0,0 +1,135 @@
+#!/bin/bash
+# install_crm_systemd.sh - 安装 CRM systemd 服务
+# Usage: sudo ./install_crm_systemd.sh [crm_dir]
+
+set -e
+
+CRM_DIR="${1:-/home/ubuntu/crm-only-20260602_18190}"
+SERVICE_NAME="sub2api-crm"
+SERVICE_FILE="/etc/systemd/system/${SERVICE_NAME}.service"
+ENV_FILE="${CRM_DIR}/.env.crm"
+
+echo "=== Installing Sub2API CRM systemd service ==="
+echo "CRM Directory: ${CRM_DIR}"
+echo "Service File: ${SERVICE_FILE}"
+
+# 检查是否为 root
+if [ "$EUID" -ne 0 ]; then
+    echo "ERROR: Please run with sudo"
+    exit 1
+fi
+
+# 检查目录存在
+if [ ! -d "${CRM_DIR}" ]; then
+    echo "ERROR: Directory ${CRM_DIR} does not exist"
+    exit 1
+fi
+
+# 检查可执行文件存在
+if [ ! -x "${CRM_DIR}/sub2api-cn-relay-manager-server" ]; then
+    echo "ERROR: Binary not found or not executable"
+    exit 1
+fi
+
+# 停止现有 nohup 进程
+echo "Stopping existing CRM processes..."
+pkill -f 'sub2api-cn-relay-manager-server' 2>/dev/null || true
+sleep 2
+
+# 从目录名提取端口
+PORT=$(echo "${CRM_DIR}" | grep -oE '[0-9]+' | tail -1)
+if [ -z "${PORT}" ]; then
+    PORT="18190"
+fi
+echo "Detected port: ${PORT}"
+
+# 获取运行用户
+RUN_USER=$(stat -c '%U' "${CRM_DIR}")
+echo "Run user: ${RUN_USER}"
+
+# 检查环境变量文件
+if [ ! -f "${ENV_FILE}" ]; then
+    echo "ERROR: Environment file ${ENV_FILE} not found"
+    exit 1
+fi
+
+# 创建 systemd service 文件
+cat > "${SERVICE_FILE}" << EOF
+[Unit]
+Description=Sub2API CRM API Server (Port ${PORT})
+After=network.target
+
+[Service]
+Type=simple
+User=${RUN_USER}
+Group=${RUN_USER}
+WorkingDirectory=${CRM_DIR}
+EnvironmentFile=${ENV_FILE}
+Environment="PATH=/usr/local/bin:/usr/bin:/bin"
+ExecStart=${CRM_DIR}/sub2api-cn-relay-manager-server
+ExecReload=/bin/kill -HUP \$MAINPID
+KillMode=process
+Restart=on-failure
+RestartSec=5
+StandardOutput=append:${CRM_DIR}/crm.log
+StandardError=append:${CRM_DIR}/crm.log
+
+# Security hardening
+NoNewPrivileges=true
+PrivateTmp=true
+ProtectSystem=strict
+ProtectHome=true
+ReadWritePaths=${CRM_DIR}
+
+[Install]
+WantedBy=multi-user.target
+EOF
+
+echo "Service file created: ${SERVICE_FILE}"
+
+# 重新加载 systemd
+systemctl daemon-reload
+
+# 启用服务
+systemctl enable "${SERVICE_NAME}"
+echo "Service enabled"
+
+# 启动服务
+echo "Starting service..."
+systemctl start "${SERVICE_NAME}"
+sleep 3
+
+# 验证服务状态
+if systemctl is-active --quiet "${SERVICE_NAME}"; then
+    echo "✓ Service is running"
+else
+    echo "ERROR: Service failed to start"
+    systemctl status "${SERVICE_NAME}" --no-pager
+    exit 1
+fi
+
+# 验证端口监听
+if ss -tlnp | grep -q ":${PORT}"; then
+    echo "✓ Port ${PORT} is listening"
+else
+    echo "WARNING: Port ${PORT} not listening"
+fi
+
+# 健康检查
+echo "Health check..."
+for i in 1 2 3; do
+    if curl -fsS "http://127.0.0.1:${PORT}/healthz" -m 2 >/dev/null 2>&1; then
+        echo "✓ Health check passed"
+        break
+    fi
+    sleep 2
+done
+
+echo ""
+echo "=== Installation complete ==="
+echo "Commands:"
+echo "  systemctl status ${SERVICE_NAME}  - View status"
+echo "  systemctl stop ${SERVICE_NAME}    - Stop service"
+echo "  systemctl start ${SERVICE_NAME}   - Start service"
+echo "  systemctl restart ${SERVICE_NAME} - Restart service"
+echo "  journalctl -u ${SERVICE_NAME} -f  - View logs"
--- a/scripts/deploy/sub2api-crm.service.template
+++ b/scripts/deploy/sub2api-crm.service.template
@@ -0,0 +1,28 @@
+[Unit]
+Description=Sub2API CRM API Server (Port 18190)
+After=network.target
+
+[Service]
+Type=simple
+User=ubuntu
+Group=ubuntu
+WorkingDirectory=/home/ubuntu/crm-only-20260602_18190
+EnvironmentFile=/home/ubuntu/crm-only-20260602_18190/.env.crm
+Environment="PATH=/usr/local/bin:/usr/bin:/bin"
+ExecStart=/home/ubuntu/crm-only-20260602_18190/sub2api-cn-relay-manager-server
+ExecReload=/bin/kill -HUP $MAINPID
+KillMode=process
+Restart=on-failure
+RestartSec=5
+StandardOutput=append:/home/ubuntu/crm-only-20260602_18190/crm.log
+StandardError=append:/home/ubuntu/crm-only-20260602_18190/crm.log
+
+# Security hardening
+NoNewPrivileges=true
+PrivateTmp=true
+ProtectSystem=strict
+ProtectHome=true
+ReadWritePaths=/home/ubuntu/crm-only-20260602_18190
+
+[Install]
+WantedBy=multi-user.target
--- a/scripts/deploy/verify_crm_deployment.sh
+++ b/scripts/deploy/verify_crm_deployment.sh
@@ -0,0 +1,129 @@
+#!/bin/bash
+# verify_crm_deployment.sh - 部署后三层验证脚本
+
+set -e
+
+CRM_DIR="${1:-/home/ubuntu/crm-only-20260602_18190}"
+PORT=$(echo "${CRM_DIR}" | grep -oE '[0-9]+' | tail -1)
+PORT="${PORT:-18190}"
+
+ERRORS=0
+
+echo "============================================"
+echo "CRM 部署三层验证"
+echo "============================================"
+echo "目录: ${CRM_DIR}"
+echo "端口: ${PORT}"
+echo ""
+
+# === Layer 1: 进程层 ===
+echo "[1/3] 进程层验证"
+
+# 检查 PID 文件
+if [ -f "${CRM_DIR}/crm.pid" ]; then
+    PID=$(cat "${CRM_DIR}/crm.pid")
+    echo "  PID 文件: ${PID}"
+    
+    # 检查进程存在
+    if ps -p "${PID}" >/dev/null 2>&1; then
+        echo "  ✓ 进程存在"
+        
+        # 检查 /proc/$PID/exe
+        EXE=$(readlink "/proc/${PID}/exe" 2>/dev/null || echo "unknown")
+        echo "  /proc/${PID}/exe: ${EXE}"
+        
+        if echo "${EXE}" | grep -q "(deleted)"; then
+            echo "  ✗ ERROR: Binary has (deleted) marker - old process still running"
+            ERRORS=$((ERRORS + 1))
+        else
+            echo "  ✓ Binary state: OK"
+        fi
+    else
+        echo "  ✗ ERROR: PID ${PID} not running"
+        ERRORS=$((ERRORS + 1))
+    fi
+else
+    echo "  ! PID 文件不存在 (可能使用 systemd)"
+    # 尝试 pgrep
+    NEW_PID=$(pgrep -f 'sub2api-cn-relay-manager-server' | head -1)
+    if [ -n "${NEW_PID}" ]; then
+        echo "  ✓ Found running process: ${NEW_PID}"
+    else
+        echo "  ✗ ERROR: No running process found"
+        ERRORS=$((ERRORS + 1))
+    fi
+fi
+
+echo ""
+
+# === Layer 2: 控制面层 ===
+echo "[2/3] 控制面层验证"
+
+# Test endpoints
+ENDPOINTS=(
+    "healthz:健康检查"
+    "api/portal/session:Portal Session"
+    "api/portal/logical-groups:Logical Groups"
+)
+
+for item in "${ENDPOINTS[@]}"; do
+    IFS=':' read -r endpoint name <<< "${item}"
+    
+    STATUS=$(curl -s "http://127.0.0.1:${PORT}/${endpoint}" -w '%{http_code}' -o /dev/null 2>/dev/null || echo "000")
+    
+    if [ "${STATUS}" = "200" ]; then
+        echo "  ✓ ${name}: HTTP ${STATUS}"
+    elif [ "${STATUS}" = "401" ] || [ "${STATUS}" = "403" ]; then
+        echo "  ~ ${name}: HTTP ${STATUS} (expected for protected endpoint)"
+    else
+        echo "  ✗ ${name}: HTTP ${STATUS}"
+        ERRORS=$((ERRORS + 1))
+    fi
+done
+
+# 检查 login_enabled
+SESSION_RESP=$(curl -s "http://127.0.0.1:${PORT}/api/portal/session" 2>/dev/null || echo "{}")
+if echo "${SESSION_RESP}" | grep -q '"login_enabled":true'; then
+    echo "  ✓ Portal login enabled"
+elif echo "${SESSION_RESP}" | grep -q '"login_enabled":false'; then
+    echo "  ✗ ERROR: Portal login disabled (SUB2API_CRM_PORTAL_SESSION_SECRET not set)"
+    ERRORS=$((ERRORS + 1))
+else
+    echo "  ! Cannot determine login_enabled state"
+fi
+
+echo ""
+
+# === Layer 3: 用户面层 ===
+echo "[3/3] 用户面层验证"
+
+# 检查 nginx 配置
+if [ -f /etc/nginx/sites-enabled/crm.sub.tksea.top.conf ] || [ -f /etc/nginx/sites-available/crm.sub.tksea.top.conf ]; then
+    echo "  ✓ Nginx config exists"
+    
+    # 检查 nginx 是否指向正确端口
+    NGINX_CONF=$(cat /etc/nginx/sites-enabled/crm.sub.tksea.top.conf 2>/dev/null || cat /etc/nginx/sites-available/crm.sub.tksea.top.conf 2>/dev/null)
+    if echo "${NGINX_CONF}" | grep -q "127.0.0.1:${PORT}"; then
+        echo "  ✓ Nginx points to port ${PORT}"
+    else
+        echo "  ! WARNING: Nginx may not point to correct port"
+    fi
+else
+    echo "  ! Nginx config not found in standard location"
+fi
+
+# 检查外部访问 (可选，可能因网络限制失败)
+# EXTERN_RESP=$(curl -s "https://crm.sub.tksea.top/api/portal/session" --connect-timeout 3 2>/dev/null || echo "") 
+# if [ -n "${EXTERN_RESP}" ]; then
+#     echo "  ✓ External access working"
+# fi
+
+echo ""
+echo "============================================"
+if [ ${ERRORS} -eq 0 ]; then
+    echo "✓ 全部验证通过"
+    exit 0
+else
+    echo "✗ 发现 ${ERRORS} 个错误"
+    exit 1
+fi
--- a/scripts/test/test_tksea_portal_assets.sh
+++ b/scripts/test/test_tksea_portal_assets.sh
@@ -14,6 +14,9 @@ ADMIN_PROVIDERS_FILE="$ROOT_DIR/deploy/tksea-portal/admin/providers.html"
 ADMIN_BATCH_FILE="$ROOT_DIR/deploy/tksea-portal/admin/batch-import.html"
 NGINX_FILE="$ROOT_DIR/deploy/tksea-portal/nginx.sub.tksea.top.conf.example"
 DEPLOY_SCRIPT="$ROOT_DIR/scripts/deploy/deploy_tksea_portal.sh"
+DEPLOY_ENV_EXAMPLE="$ROOT_DIR/scripts/deploy/.env.deploy.example"
+DEPLOY_CRM_ONLY_SCRIPT="$ROOT_DIR/scripts/deploy/deploy_crm_only.sh"
+REMOTE43_SETUP_SCRIPT="$ROOT_DIR/scripts/deploy/setup_remote43_patched_stack.sh"

 fail() {
  echo "FAIL: $*" >&2
@@ -28,6 +31,14 @@ assert_contains_file() {
  fi
 }

+assert_not_contains_file() {
+  local file="$1"
+  local needle="$2"
+  if grep -Fq "$needle" "$file"; then
+    fail "did not expect [$needle] in $file"
+  fi
+}
+
 [[ -f "$HTML_FILE" ]] || fail "missing $HTML_FILE"
 [[ -f "$ADMIN_HTML_FILE" ]] || fail "missing $ADMIN_HTML_FILE"
 [[ -f "$ADMIN_COMMON_CSS_FILE" ]] || fail "missing $ADMIN_COMMON_CSS_FILE"
@@ -40,6 +51,9 @@ assert_contains_file() {
 [[ -f "$ADMIN_BATCH_FILE" ]] || fail "missing $ADMIN_BATCH_FILE"
 [[ -f "$NGINX_FILE" ]] || fail "missing $NGINX_FILE"
 [[ -f "$DEPLOY_SCRIPT" ]] || fail "missing $DEPLOY_SCRIPT"
+[[ -f "$DEPLOY_ENV_EXAMPLE" ]] || fail "missing $DEPLOY_ENV_EXAMPLE"
+[[ -f "$DEPLOY_CRM_ONLY_SCRIPT" ]] || fail "missing $DEPLOY_CRM_ONLY_SCRIPT"
+[[ -f "$REMOTE43_SETUP_SCRIPT" ]] || fail "missing $REMOTE43_SETUP_SCRIPT"

 assert_contains_file "$HTML_FILE" "Sub2API 多模型接入中心"
 assert_contains_file "$HTML_FILE" "https://sub.tksea.top/portal/"
@@ -111,7 +125,24 @@ assert_contains_file "$ADMIN_COMMON_JS_FILE" "/portal/admin/route-health.html"
 assert_contains_file "$ADMIN_COMMON_JS_FILE" "/portal/admin/accounts.html"
 assert_contains_file "$ADMIN_COMMON_JS_FILE" "/portal/admin/providers.html"
 assert_contains_file "$ADMIN_COMMON_JS_FILE" "/portal/admin/batch-import.html"
+# 契约修正：crm_subject 是 trusted subject 来源，crm_session 已弃用
+assert_contains_file "$NGINX_FILE" '$cookie_crm_subject'
+assert_not_contains_file "$NGINX_FILE" '$cookie_crm_session'
+assert_contains_file "$DEPLOY_SCRIPT" '$cookie_crm_subject'
+assert_not_contains_file "$DEPLOY_SCRIPT" '$cookie_crm_session'
 assert_contains_file "$ADMIN_COMMON_JS_FILE" "/portal/"
+assert_contains_file "$ADMIN_COMMON_JS_FILE" "filterSensitiveData"
+assert_contains_file "$ADMIN_COMMON_JS_FILE" "containsSensitiveData(payload)"
+
+assert_not_contains_file "$ADMIN_ACCOUNTS_FILE" "adminToken: adminTokenInput.value,"
+assert_not_contains_file "$ADMIN_PROVIDERS_FILE" "adminToken: adminTokenInput.value,"
+assert_not_contains_file "$ADMIN_PROVIDERS_FILE" "accessAPIKey: accessAPIKeyInput.value.trim(),"
+assert_not_contains_file "$ADMIN_PROVIDERS_FILE" "providerKeys: providerKeysInput.value,"
+assert_not_contains_file "$ADMIN_HTML_FILE" "adminToken: adminTokenInput.value,"
+assert_not_contains_file "$ADMIN_HTML_FILE" "probeAPIKey: probeAPIKeyInput.value.trim(),"
+assert_not_contains_file "$ADMIN_HTML_FILE" "entries: entriesInput.value,"
+assert_not_contains_file "$ADMIN_ROUTE_HEALTH_FILE" "localStorage.setItem(storageKey"
+assert_not_contains_file "$ADMIN_LOGICAL_GROUPS_FILE" "localStorage.setItem(storageKey"

 assert_contains_file "$ADMIN_HTML_FILE" "Batch Import Admin"
 assert_contains_file "$ADMIN_HTML_FILE" "/portal/admin-common.css"
@@ -190,6 +221,12 @@ assert_contains_file "$ADMIN_ACCOUNTS_FILE" "/portal-admin-api"
 assert_contains_file "$ADMIN_PROVIDERS_FILE" "Provider Admin"
 assert_contains_file "$ADMIN_PROVIDERS_FILE" "/portal/admin-common.css"
 assert_contains_file "$ADMIN_PROVIDERS_FILE" "/portal/admin-common.js"
+assert_not_contains_file "$DEPLOY_CRM_ONLY_SCRIPT" 'KEY="${KEY:-/home/long/下载/zjsea.pem}"'
+assert_not_contains_file "$DEPLOY_CRM_ONLY_SCRIPT" 'REMOTE="${REMOTE:-ubuntu@43.155.133.187}"'
+assert_not_contains_file "$REMOTE43_SETUP_SCRIPT" 'KEY="${KEY:-/home/long/下载/zjsea.pem}"'
+assert_not_contains_file "$REMOTE43_SETUP_SCRIPT" 'REMOTE="${REMOTE:-ubuntu@43.155.133.187}"'
+assert_contains_file "$DEPLOY_CRM_ONLY_SCRIPT" '.env.deploy.example'
+assert_contains_file "$REMOTE43_SETUP_SCRIPT" '.env.deploy.example'
 assert_contains_file "$ADMIN_PROVIDERS_FILE" "data-admin-nav"
 assert_contains_file "$ADMIN_PROVIDERS_FILE" "管理员登录"
 assert_contains_file "$ADMIN_PROVIDERS_FILE" "/api/packs"
@@ -237,5 +274,12 @@ assert_contains_file "$DEPLOY_SCRIPT" "REMOTE_PORTAL_DIR"
 assert_contains_file "$DEPLOY_SCRIPT" "REMOTE_CRM_PORT"
 assert_contains_file "$DEPLOY_SCRIPT" "LOCAL_PORTAL_DIR"
 assert_contains_file "$DEPLOY_SCRIPT" "patch_tksea_portal_nginx.py"
+assert_contains_file "$DEPLOY_SCRIPT" "DEPLOY_ENV_FILE"
+assert_not_contains_file "$DEPLOY_SCRIPT" "/home/long/下载/zjsea.pem"
+assert_not_contains_file "$DEPLOY_SCRIPT" "ubuntu@43.155.133.187"
+
+assert_contains_file "$DEPLOY_ENV_EXAMPLE" "KEY="
+assert_contains_file "$DEPLOY_ENV_EXAMPLE" "REMOTE="
+assert_contains_file "$DEPLOY_ENV_EXAMPLE" "REMOTE_CRM_PORT="

 echo "PASS: tksea portal assets look consistent"