From e1e0c05d3a3493059366d395b78aabae86aa8e78 Mon Sep 17 00:00:00 2001
From: "sanze.li" <weixin.li@nio.com>
Date: Sat, 29 Aug 2026 23:04:10 +0800
Subject: [PATCH 1/3] feat: add Cursor Agent CLI managed entry

---
 .sopify/blueprint/README.md                   |  3 +-
 .sopify/blueprint/design.md                   |  4 +-
 .../20260828_cursor_agent_cli_entry/plan.md   | 79 +++++++++++++++++++
 .../receipts/exec_001.json                    | 18 +++++
 .../receipts/exec_002.json                    | 25 ++++++
 .../receipts/exec_003.json                    | 41 ++++++++++
 .../receipts/exec_004.json                    | 39 +++++++++
 .../receipts/exec_005.json                    | 21 +++++
 .../receipts/verify_001.json                  | 40 ++++++++++
 .../receipts/verify_002.json                  | 31 ++++++++
 .../receipts/verify_003.json                  | 45 +++++++++++
 .../20260828_cursor_agent_cli_entry/tasks.md  | 32 ++++++++
 CHANGELOG.md                                  | 13 +++
 README.md                                     |  2 +-
 README.zh-CN.md                               |  2 +-
 docs/cursor-host.md                           | 15 ++--
 docs/getting-started.md                       | 17 ++--
 docs/how-sopify-works.en.md                   |  2 +-
 docs/how-sopify-works.md                      |  2 +-
 index.html                                    |  2 +-
 installer/cursor_plugin.py                    |  9 +++
 installer/hosts/base.py                       |  1 +
 installer/inspection.py                       |  4 +-
 skills/en/cursor-cli-skill.md.template        | 23 ++++++
 skills/en/cursor-plugin-readme.md.template    |  9 ++-
 skills/en/cursor-plugin-rule.mdc.template     |  7 +-
 skills/zh/cursor-cli-skill.md.template        | 23 ++++++
 skills/zh/cursor-plugin-readme.md.template    |  9 ++-
 skills/zh/cursor-plugin-rule.mdc.template     |  7 +-
 tests/test_installer_status_doctor.py         | 57 ++++++++++++-
 tests/test_plan_contract_assets.py            |  4 +-
 zh-CN.html                                    |  2 +-
 32 files changed, 546 insertions(+), 42 deletions(-)
 create mode 100644 .sopify/plan/20260828_cursor_agent_cli_entry/plan.md
 create mode 100644 .sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_001.json
 create mode 100644 .sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_002.json
 create mode 100644 .sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_003.json
 create mode 100644 .sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_004.json
 create mode 100644 .sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_005.json
 create mode 100644 .sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_001.json
 create mode 100644 .sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_002.json
 create mode 100644 .sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_003.json
 create mode 100644 .sopify/plan/20260828_cursor_agent_cli_entry/tasks.md
 create mode 100644 skills/en/cursor-cli-skill.md.template
 create mode 100644 skills/zh/cursor-cli-skill.md.template

diff --git a/.sopify/blueprint/README.md b/.sopify/blueprint/README.md
index 0b50622..0fe07a5 100644
--- a/.sopify/blueprint/README.md
+++ b/.sopify/blueprint/README.md
@@ -13,7 +13,7 @@
 ## 当前焦点
 
 <!-- sopify:auto:focus:start -->
-- 当前活动 plan：无；`20260720_plain_language_output` 已完成 Skill 表达、产品页和图文收口，并通过 Sopify `2026-07-21` Release 与 `main / root` GitHub Pages 完成公开交付。
+- 当前活动 plan：`../plan/20260828_cursor_agent_cli_entry` 已完成 Cursor Agent CLI managed-first 顶层入口与一次安装产品化，本机安装和两轮 Cursor 独立审计通过，状态为 `ready_to_archive`；feat 分支 commit/push 已获授权，尚未 finalize、合并或发布。
 - MCP tool plane 已完成 Codex-first 持久注册验证和 Claude 会话级只读调用验证，Qoder / Claude / Copilot 的后续持久注册仍按证据推进。
 - P8 Protocol Kernel & Runtime Retirement 已归档至 `../history/2026-06/20260605_p8_protocol_kernel_runtime_retirement`。
 - history 归档：已可用；最近归档：`../history/2026-07/20260720_plain_language_output`。
@@ -28,6 +28,7 @@
 - [蓝图任务](./tasks.md)
 - [Sopify 宿主接入规范 (Protocol v0)](./protocol.md)
 - [Skill 标准对齐蓝图](./skill-standards-refactor.md)
+- [Cursor Agent CLI 一次安装方案](../plan/20260828_cursor_agent_cli_entry/plan.md)
 - [变更历史](../history/index.md)
 - 最近归档：`../history/2026-07/20260720_plain_language_output`
 <!-- sopify:auto:read-next:end -->
diff --git a/.sopify/blueprint/design.md b/.sopify/blueprint/design.md
index 4541384..e7c4e2b 100644
--- a/.sopify/blueprint/design.md
+++ b/.sopify/blueprint/design.md
@@ -441,7 +441,7 @@ P4b 减重和 P4c 宿主消费治理的红线边界。只冻结 artifact / schem
 
 > **未列入面默认可删**：`state/sessions/*`、`state/last_route.json`、runtime 内部模块边界、route name 全集、output 渲染文案措辞均为 runtime 内部实现，不在 keep-list 内。P4b 减重时可自由处置。
 
-`host_support` 只声明官方适配器能把 Skill 语义交付到对应宿主的支持界面并由宿主消费。它不等同于原生 Skill discovery，也不证明该宿主已经完成 E2E；支持等级、已验证能力和入口形态继续由 `HostCapability` 表达。当前五个内置 Skill 面向 `codex / claude / qoder / copilot / cursor`，其中 Copilot 通过单文件展开消费，Cursor IDE 通过用户 Plugin Rule 与全局 Skill 树拆分消费；Agent CLI 只保留手工 Skill 与用户 Hook 兼容面。
+`host_support` 只声明官方适配器能把 Skill 语义交付到对应宿主的支持界面并由宿主消费。它不等同于原生 Skill discovery，也不证明该宿主已经完成 E2E；支持等级、已验证能力和入口形态继续由 `HostCapability` 表达。当前五个内置阶段 Skill 面向 `codex / claude / qoder / copilot / cursor`，其中 Copilot 通过单文件展开消费；Cursor IDE 通过用户 Plugin Rule 进入，本地 Agent CLI 默认安装 managed-first 顶层 Skill，选中后读取同一 Rule，并与 IDE 共用阶段 Skill 树、payload 与用户 Hooks。CLI 模型选择属于 best-effort，安装事实不替代行为证据。
 
 
 ## Output Rendering Audit — *[pre-P8 legacy reference]*
@@ -504,7 +504,7 @@ output.py 渲染层逐字段分类。只做分类，不做改造决策（改造
 | Claude | `protocol_verified` | `install.sh --target claude:zh-CN` | CONTINUATION + INTERACTION + AUDIT | 已验证 |
 | Qoder | `protocol_verified` | `install.sh --target qoder` | CONTINUATION + INTERACTION + AUDIT | W3.1-W3.3 已验证 |
 | Copilot | `baseline_supported` | `install.sh --target copilot` | PROMPT_ONLY | Prompt-only |
-| Cursor | `baseline_supported` | `install.sh --target cursor` | PROMPT_ONLY + HOOKS | IDE user Plugin Rule + global Skills/payload + user-level `~/.cursor/hooks.json`; observable IDE baseline evidence, CLI manual Skills only |
+| Cursor | `baseline_supported` | `install.sh --target cursor` | PROMPT_ONLY + HOOKS | IDE user Plugin Rule + managed-first CLI entry + shared Skills/payload + user-level `~/.cursor/hooks.json`; IDE/CLI behavior evidence remains separate |
 
 ### 契约消费矩阵（P8 2-file state model）
 
diff --git a/.sopify/plan/20260828_cursor_agent_cli_entry/plan.md b/.sopify/plan/20260828_cursor_agent_cli_entry/plan.md
new file mode 100644
index 0000000..c2ea567
--- /dev/null
+++ b/.sopify/plan/20260828_cursor_agent_cli_entry/plan.md
@@ -0,0 +1,79 @@
+---
+title: Cursor Agent CLI 顶层入口与一次安装产品化
+plan_id: 20260828_cursor_agent_cli_entry
+status: done
+lifecycle_state: ready_to_archive
+level: standard
+created: 2026-08-28
+updated: 2026-08-29
+archive_ready: true
+---
+
+# Cursor Agent CLI 顶层入口与一次安装产品化
+
+就绪状态: Ready
+依据: 用户已确认 managed-first 口径，并授权先优化本机 CLI 顶层 Skill、安装最新版、经 Cursor 独立审计后，再纳入现有 Cursor 一次安装产品链；不新增 launcher、项目 Rule、runtime 或独立 CLI target。
+
+## Context / Why
+
+Cursor IDE 已通过用户级 Always Plugin Rule 获得 Sopify 顶层协议。2026-08-28 的本机实验进一步证明 `agent` CLI 可以发现 `~/.cursor/skills/sopify/SKILL.md`，并通过该薄入口读取已安装 Plugin Rule、路由五个阶段 Skill。
+
+受控实验不等于稳定产品能力。安装后的三个真实用户会话中，顶层 Skill 只自动命中一次；命中会话还出现了入口 Skill 与业务素材并行读取，以及 quick-fix 从默认应用修改扩展到缓存删除和 Helper App 持久化、未重新停车的问题。当前顶层 Skill 只有用户目录单份副本，仓库安装源、升级保护、Doctor、双语文档与产品声明均未纳入；重新运行现有 Cursor 安装会删除这份本机入口。
+
+## Scope
+
+- 第一层：按真实使用证据优化本机中文 CLI 顶层 Skill，清理旧副本并安装唯一最新版；保持入口短小、结构清晰、语言通顺。
+- 第一层验证：校验 Skill 结构与安装哈希，完成 managed 请求前向测试，然后交给 Cursor 在独立标签审计；不认可则按审计结论做最小修正并复验。
+- 第二层：将中英文 CLI 顶层 Skill 纳入 canonical source，使现有 `cursor:zh-CN` / `cursor:en-US` 一次安装默认同时覆盖 IDE 与本地 Agent CLI。
+- 第二层验证：补安装完整性、升级/重装、Doctor、IDE 回归和 CLI managed 流程证据；实现完成后再次交给 Cursor 独立审计。
+- 同步受影响的 Cursor 接入文档、双语产品说明、Changelog 与稳定 blueprint 事实；保持支持档位和未验证边界准确。
+
+## Approach
+
+保留“薄 CLI 入口 → 读取已安装 Plugin Rule → 路由现有阶段 Skills”的结构。顶层 Skill 的 description 从“任何软件请求”收窄为 managed-first 高信号触发，突出 `.sopify`、`~go`、分析/设计/开发、继续/接续/恢复和 finalize/收口；普通咨询不强制触发，也不据此声称已进入 Sopify。
+
+Plugin Rule 只补一个共享边界：当已授权动作扩展为新的用户级或系统级持久化、删除操作时重新停车确认。产品化复用现有 Cursor 用户级安装与 `copytree`，把顶层 Skill 加入 canonical 双语源和安装必需路径，不增加新 target、安装参数、能力枚举或运行时组件。
+
+## Waves / Steps
+
+- [x] W1：完成本机中文顶层 Skill 受控实验，并保存旧版本 receipts。
+- [x] W2：根据近期真实会话优化本机 Skill，安装唯一最新版并完成 Cursor 独立审计。
+- [x] W3：把中英文入口纳入一次安装、Doctor/测试和用户文档，保持 IDE/CLI 证据分离。
+- [x] W4：执行目标测试、全量回归、CLI managed 前向观察和第二次 Cursor 独立审计；IDE/CLI 行为证据继续分开。
+- [x] W5：按验证结论同步稳定知识并更新为 `ready_to_archive`；不自动 finalize、commit、push 或发布。
+
+## Key Decisions
+
+- 采用 managed-first：只保证 Sopify/managed 请求的顶层入口，大部分相关场景应自然触发；普通咨询不强制。
+- 用户仍只运行一次 `--target cursor:<language>`；IDE 与本地 CLI 共用阶段 Skills、payload 与 Hooks，但入口机制和行为证据分开声明。
+- 本机 Skill 与产品源使用同一正文，不保留旧版、兼容副本或额外路由文件。
+- 自动 Skill 选择按 best-effort 声明；安装存在、用户已读已用和模型自述均不单独证明稳定行为。
+- Cursor 保持 `BASELINE_SUPPORTED`；Cloud Agent、跨模型一致性和 `PROTOCOL_VERIFIED` 升级不在本轮承诺内。
+
+## Constraints / Not-in-scope
+
+- 不新增 `cursor-cli` target、`--with-cursor-cli`、launcher、项目 `.cursor/rules`、`AGENTS.md`、Custom Mode、MCP 或 runtime。
+- 不修改 Cursor settings、CLI config、代理、模型、账号、API Key、钥匙串或 Cloud Agent。
+- 不重构五项阶段 Skill、writer、state schema、Hook 体系或 HostCapability 架构。
+- 不把通用问答未命中视为失败；只对 managed 触发、读取顺序、协议消费和副作用边界做验收。
+- 不保存含敏感内容的完整 transcript；正式证据记录 session id、受限 tool trace 摘要、用户观察和 digest。
+
+knowledge_sync:
+  project: skip
+  background: skip
+  design: required
+  tasks: skip
+
+## Status / Progress
+
+- [x] 本机旧 Skill、受控 receipts、真实会话命中率和现有安装覆盖行为已完成批判审计。
+- [x] 用户确认 managed-first、一次安装默认支持 IDE + CLI、不过度设计，并要求两阶段均经 Cursor 独立审计。
+- [x] 第一层本机 Skill 优化与审计：当前唯一 Skill SHA-256 为 `d892a90fdb7c455367d870e5b47f454a422d578109919d93c9a68eb5fd0beee9`；Cursor 首审不认可读取顺序只在正文，修正 description 后复审认可。
+- [x] 第二层产品化实现：中英文顶层 Skill、一次安装内容校验、旧入口清理、Doctor 必需路径、共享 Rule 边界和用户文档已完成；目标测试与全量 `312 passed / 78 subtests` 通过。
+- [x] 跨宿主污染审计与修正：Cursor 首审不认可专用入口进入共享 Skill 树；现已改为 host-specific 双语模板，补 Codex/Claude/Qoder 负向测试并恢复 shared-tree golden。
+- [x] 第二层 Cursor 最终复审认可，无待修项；最终全量回归为 `313 passed / 81 subtests`。
+- [x] 用户已授权在 `feat/cursor-agent-cli-entry` 本地 commit 并 push 同名 feat 分支；不包含 finalize、合并或发布。
+
+## Next
+
+实现与独立审计均已完成。按用户授权将当前结果 commit 并 push 到 `feat/cursor-agent-cli-entry`；只有显式 `~go finalize` 才归档，合并与发布继续需要单独授权。
diff --git a/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_001.json b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_001.json
new file mode 100644
index 0000000..94f0075
--- /dev/null
+++ b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_001.json
@@ -0,0 +1,18 @@
+{
+  "evidence": {
+    "baseline_greeting_sha256": "5891b5b522d5df086d0ff0b110fbd9d21bb4fc7163af34d08286a2e846f6be03",
+    "demo_root": "/private/tmp/sopify-cursor-cli-entry-demo-20260828",
+    "installed_skill_sha256": "bbec80a93fd1acf049e7a5539930d7c93337e60dbfb97ce356f48fd4e9ffdafd",
+    "product_source_diff": "none",
+    "repository_regression": "311 passed, 78 subtests"
+  },
+  "provenance": {
+    "plan_id": "20260828_cursor_agent_cli_entry",
+    "plan_version": "sha256:576bf51018ab71222cacb7e8e4d7b048afa04eb3c6ba60908be4b33a3fce4b8b",
+    "receipt_id": "exec_001",
+    "scope": "local Cursor Agent CLI entry probe",
+    "writer": "sopify_writer"
+  },
+  "timestamp": "2026-08-28T13:39:48+00:00",
+  "verdict": "local_probe_ready"
+}
diff --git a/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_002.json b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_002.json
new file mode 100644
index 0000000..74a1aee
--- /dev/null
+++ b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_002.json
@@ -0,0 +1,25 @@
+{
+  "evidence": {
+    "cli_config_note": "Cursor Agent later refreshed its own dynamic CLI cache; this is not attributed to the Sopify installer.",
+    "cli_config_sha256_before_and_immediately_after_installer": "406ef6519b0fbd8bca8f8e36ea334c171833f91a7cd36aa6a31c0227b7ec496e",
+    "cursor_install_checks": {
+      "cursor_hooks_present": "pass",
+      "cursor_plugin_present": "pass",
+      "global_skill_tree_present": "pass",
+      "payload_present": "pass"
+    },
+    "installed_skill_sha256": "d892a90fdb7c455367d870e5b47f454a422d578109919d93c9a68eb5fd0beee9",
+    "installed_sopify_version": "2026-08-22.134737",
+    "skill_validation": "passed",
+    "stale_top_level_skill_copies": 0
+  },
+  "provenance": {
+    "host": "codex",
+    "plan_id": "20260828_cursor_agent_cli_entry",
+    "plan_version": "sha256:11479834b277811893455f73d63e30144d99ac34a9ec0e7bfa5a012ef0052208",
+    "receipt_id": "exec_002",
+    "scope": "local Cursor CLI Skill update"
+  },
+  "timestamp": "2026-08-29T14:30:38+00:00",
+  "verdict": "local_cli_skill_updated"
+}
diff --git a/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_003.json b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_003.json
new file mode 100644
index 0000000..eb75256
--- /dev/null
+++ b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_003.json
@@ -0,0 +1,41 @@
+{
+  "evidence": {
+    "canonical_entry_paths": [
+      "skills/zh/skills/sopify/SKILL.md",
+      "skills/en/skills/sopify/SKILL.md"
+    ],
+    "cli_config_installer_observation": "unchanged across the final product install; Cursor Agent itself updates dynamic cache fields during sessions",
+    "installed_cursor_checks": {
+      "cursor_cli_behavior": "BLACK_BOX_NOT_VERIFIED",
+      "cursor_hooks_present": "pass",
+      "cursor_ide_behavior": "BLACK_BOX_NOT_VERIFIED",
+      "cursor_plugin_present": "pass",
+      "global_skill_tree_present": "pass",
+      "payload_present": "pass"
+    },
+    "installed_skill_sha256": "d892a90fdb7c455367d870e5b47f454a422d578109919d93c9a68eb5fd0beee9",
+    "installer_contract": {
+      "new_target_or_flag": false,
+      "repeat_install_action": "skipped",
+      "root_skill_content_verified": true,
+      "stale_local_entry_replaced": true,
+      "target": "cursor:<language>"
+    },
+    "tests": {
+      "full": "312 passed, 78 subtests",
+      "git_diff_check": "passed",
+      "golden_snapshots": "24 passed",
+      "targeted": "93 passed, 27 subtests",
+      "version_consistency": "passed"
+    }
+  },
+  "provenance": {
+    "host": "codex",
+    "plan_id": "20260828_cursor_agent_cli_entry",
+    "plan_version": "sha256:1dd10d17b6cba52a06ae75d3709e34103b25b187a46107e00b3372495cd3a592",
+    "receipt_id": "exec_003",
+    "scope": "Cursor IDE and local Agent CLI one-install productization"
+  },
+  "timestamp": "2026-08-29T14:38:42+00:00",
+  "verdict": "cursor_cli_one_install_productized"
+}
diff --git a/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_004.json b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_004.json
new file mode 100644
index 0000000..4a20f43
--- /dev/null
+++ b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_004.json
@@ -0,0 +1,39 @@
+{
+  "evidence": {
+    "correction": {
+      "cursor_installer_writes_root_skill_separately": true,
+      "en_source": "skills/en/cursor-cli-skill.md.template",
+      "non_cursor_negative_hosts": [
+        "codex",
+        "claude",
+        "qoder"
+      ],
+      "shared_tree_golden_restored": true,
+      "shared_tree_root_skill_present": false,
+      "zh_source": "skills/zh/cursor-cli-skill.md.template"
+    },
+    "cursor_audit": {
+      "dispatch_id": "ctx_750e36ccddfc",
+      "finding": "Cursor-only root Skill was placed in the shared host Skill tree",
+      "message_id": "msg_63e9bc828252",
+      "verdict": "not_accepted"
+    },
+    "installed_skill_sha256": "d892a90fdb7c455367d870e5b47f454a422d578109919d93c9a68eb5fd0beee9",
+    "repeat_install_action": "skipped",
+    "tests": {
+      "focused": "73 passed, 30 subtests",
+      "full": "313 passed, 81 subtests",
+      "git_diff_check": "passed"
+    }
+  },
+  "provenance": {
+    "host": "codex",
+    "plan_id": "20260828_cursor_agent_cli_entry",
+    "plan_version": "sha256:1dd10d17b6cba52a06ae75d3709e34103b25b187a46107e00b3372495cd3a592",
+    "receipt_id": "exec_004",
+    "scope": "cross-host isolation correction",
+    "verifier_host": "cursor"
+  },
+  "timestamp": "2026-08-29T14:49:30+00:00",
+  "verdict": "cursor_cli_entry_isolated_from_shared_host_tree"
+}
diff --git a/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_005.json b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_005.json
new file mode 100644
index 0000000..3c0854f
--- /dev/null
+++ b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/exec_005.json
@@ -0,0 +1,21 @@
+{
+  "evidence": {
+    "branch": "feat/cursor-agent-cli-entry",
+    "commit_authorized": true,
+    "cursor_final_review": "accepted with no remaining findings",
+    "finalize_authorized": false,
+    "last_full_test_result": "313 passed, 81 subtests",
+    "merge_authorized": false,
+    "push_authorized": true,
+    "release_authorized": false
+  },
+  "provenance": {
+    "host": "codex",
+    "plan_id": "20260828_cursor_agent_cli_entry",
+    "plan_version": "sha256:9a481e9161773c7d3ccb43aba20c0242e7e6dc85fc9a1861d93abdaf8746bf40",
+    "receipt_id": "exec_005",
+    "scope": "local commit and feature-branch push authorization"
+  },
+  "timestamp": "2026-08-29T15:03:13+00:00",
+  "verdict": "feature_branch_delivery_authorized"
+}
diff --git a/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_001.json b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_001.json
new file mode 100644
index 0000000..a18fc06
--- /dev/null
+++ b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_001.json
@@ -0,0 +1,40 @@
+{
+  "evidence": {
+    "cases": {
+      "analyze_design": "pass",
+      "consult": "pass",
+      "develop": "pass"
+    },
+    "installed_skill_sha256": "bbec80a93fd1acf049e7a5539930d7c93337e60dbfb97ce356f48fd4e9ffdafd",
+    "product_source_diff": "none",
+    "repository_regression": "311 passed, 78 subtests",
+    "transcript_sha256": {
+      "analyze_design": "26b0f7c2c1765f4714fc543c9f01cf2c2461861db97cfd9b2b44a786cdb1a293",
+      "consult": "a7eb73fa5db21e7fcd3c3ea31e86e590e684dd66093964328ab7cbe9a371e5d2",
+      "develop": "e7932475f303926bdaf88c5191f2c24396896abd05cec825aff294f238c73bec"
+    },
+    "unverified_boundaries": [
+      "Cursor IDE behavior after local root Skill install",
+      "cross-model stability",
+      "bilingual product installation"
+    ],
+    "verified_boundaries": [
+      "automatic root Skill selection in three fresh Cursor CLI sessions",
+      "consult read-only",
+      "Analyze score and Design level scripts",
+      "four-step continuation",
+      "sopify_writer state/handoff/receipts",
+      "state ignore",
+      "ready_to_archive without commit/finalize"
+    ]
+  },
+  "provenance": {
+    "host": "codex",
+    "plan_id": "20260828_cursor_agent_cli_entry",
+    "plan_version": "sha256:6060df26c8bfdae38577e4020d234a36dac2f1544b86f20e6e4639503608220b",
+    "receipt_id": "verify_001",
+    "scope": "local Cursor Agent CLI entry experiment"
+  },
+  "timestamp": "2026-08-28T14:46:07+00:00",
+  "verdict": "accept_local_cli_entry"
+}
diff --git a/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_002.json b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_002.json
new file mode 100644
index 0000000..ead010e
--- /dev/null
+++ b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_002.json
@@ -0,0 +1,31 @@
+{
+  "evidence": {
+    "correction": "moved the Skill and Plugin Rule first-read constraint into the frontmatter description",
+    "first_cursor_audit": {
+      "dispatch_id": "ctx_2ffacf6dfec5",
+      "finding": "initial read-order constraint was absent from description",
+      "verdict": "not_accepted"
+    },
+    "forward_observation": [
+      "Cursor selected the sopify root Skill for the managed audit request",
+      "the installed Plugin Rule was read before business execution",
+      "the reviewing Cursor worker made no task-file mutations"
+    ],
+    "installed_skill_sha256": "d892a90fdb7c455367d870e5b47f454a422d578109919d93c9a68eb5fd0beee9",
+    "second_cursor_audit": {
+      "dispatch_id": "ctx_138ea9a1430b",
+      "message_id": "msg_d5529bd8e813",
+      "verdict": "accepted"
+    }
+  },
+  "provenance": {
+    "host": "codex",
+    "plan_id": "20260828_cursor_agent_cli_entry",
+    "plan_version": "sha256:11479834b277811893455f73d63e30144d99ac34a9ec0e7bfa5a012ef0052208",
+    "receipt_id": "verify_002",
+    "scope": "local Cursor CLI Skill independent review",
+    "verifier_host": "cursor"
+  },
+  "timestamp": "2026-08-29T14:30:38+00:00",
+  "verdict": "accept_local_cli_skill_after_cursor_review"
+}
diff --git a/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_003.json b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_003.json
new file mode 100644
index 0000000..bb834cf
--- /dev/null
+++ b/.sopify/plan/20260828_cursor_agent_cli_entry/receipts/verify_003.json
@@ -0,0 +1,45 @@
+{
+  "evidence": {
+    "code_quality": "passed: host-specific source, content-aware reinstall, minimal negative coverage, no new target or runtime surface",
+    "cursor_reviews": [
+      {
+        "dispatch_id": "ctx_750e36ccddfc",
+        "finding": "Cursor-only root Skill polluted the shared host tree",
+        "verdict": "not_accepted"
+      },
+      {
+        "dispatch_id": "ctx_d2cb261a215a",
+        "finding": "cross-host isolation, minimal tests, and documentation are closed",
+        "message_id": "msg_ca3e063cf5bd",
+        "verdict": "accepted"
+      }
+    ],
+    "final_architecture": {
+      "cursor_only_sources": [
+        "skills/zh/cursor-cli-skill.md.template",
+        "skills/en/cursor-cli-skill.md.template"
+      ],
+      "installed_path": "~/.cursor/skills/sopify/SKILL.md",
+      "shared_host_tree_pollution": false,
+      "single_target": "cursor:<language>",
+      "support_tier": "BASELINE_SUPPORTED"
+    },
+    "final_tests": "313 passed, 81 subtests",
+    "remaining_boundaries": [
+      "IDE and CLI behavior checks remain BLACK_BOX_NOT_VERIFIED",
+      "Cloud Agent excluded",
+      "cross-model selection not promised"
+    ],
+    "spec_compliance": "passed: managed-first, one install, local latest, stale cleanup, independent Cursor acceptance, no finalize"
+  },
+  "provenance": {
+    "host": "codex",
+    "plan_id": "20260828_cursor_agent_cli_entry",
+    "plan_version": "sha256:26d94ce48ae59cb572a5f3bc67a234a99a681786a49fda072a989b18059d77a6",
+    "receipt_id": "verify_003",
+    "scope": "final Cursor CLI productization review",
+    "verifier_host": "cursor"
+  },
+  "timestamp": "2026-08-29T14:57:12+00:00",
+  "verdict": "accept_cursor_cli_one_install_productization"
+}
diff --git a/.sopify/plan/20260828_cursor_agent_cli_entry/tasks.md b/.sopify/plan/20260828_cursor_agent_cli_entry/tasks.md
new file mode 100644
index 0000000..f97401f
--- /dev/null
+++ b/.sopify/plan/20260828_cursor_agent_cli_entry/tasks.md
@@ -0,0 +1,32 @@
+# 任务清单: Cursor Agent CLI 顶层入口与一次安装产品化
+
+目录: `.sopify/plan/20260828_cursor_agent_cli_entry/`
+
+## 1. 本机 CLI Skill 优化
+
+- [x] 1.1 根据近期真实会话收窄 managed-first description，保持薄入口结构并补充必要读取顺序。
+- [x] 1.2 校验 Skill 结构、语言和行为边界，清理旧副本并把唯一最新版安装到 `~/.cursor/skills/sopify/SKILL.md`。
+- [x] 1.3 执行 managed 请求前向验证，记录可复核的本机证据。
+
+## 2. 第一轮 Cursor 独立审计
+
+- [x] 2.1 将当前 Skill、证据与审计要求放入 worktree 可读 handoff。
+- [x] 2.2 启动 Cursor 独立审计；收回认可/不认可结论并释放 worker 资源。
+- [x] 2.3 对不认可项做最小修正并复验，直到本机 Skill 获得认可。
+
+## 3. 一次安装产品化
+
+- [x] 3.1 将中英文顶层 Skill 纳入 canonical source，并让 Cursor 安装完整性包含该入口。
+- [x] 3.2 补最小共享副作用边界、升级/重装与 Doctor 测试；不新增 target、launcher 或能力框架。
+- [x] 3.3 同步 Cursor Plugin 说明、接入文档、双语 README、Changelog 与 blueprint 稳定事实。
+
+## 4. 产品化验证与独立审计
+
+- [x] 4.1 运行目标测试、全量回归、Git diff 检查以及安装前后受保护文件哈希检查。
+- [x] 4.2 执行 CLI managed 前向观察并保留可复核证据；IDE/CLI 行为项继续明确为独立 `BLACK_BOX_NOT_VERIFIED`，不以静态安装冒充行为证明。
+- [x] 4.3 交给 Cursor 独立审计完整实现；修正跨宿主污染后复审获得认可。
+
+## 5. 收尾
+
+- [x] 5.1 完成 spec compliance 与 code quality 两阶段复审，更新方案和任务状态。
+- [x] 5.2 将方案更新为 `ready_to_archive` 并保留在 `plan/`；不自动 finalize、commit、push 或发布。
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 13302a2..7db1764 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,6 +6,19 @@ Format: Summary → Changed → Plan Packages. File-level details live in `git l
 
 ## [Unreleased]
 
+### Summary
+
+Extended the existing Cursor user-scope installation with a managed-first entry for local Agent CLI.
+
+### Changed
+
+- **Cursor Agent CLI**: Added bilingual top-level Skills that read the installed Plugin Rule for managed requests, kept automatic model selection best-effort, and preserved separate IDE/CLI behavior evidence.
+- **Cursor installation**: The existing `cursor:<language>` target now installs and verifies the CLI entry, replaces stale local copies on reinstall, and keeps settings, proxy, model, account, and Cloud Agent out of scope.
+
+### Plan Packages
+
+- `20260828_cursor_agent_cli_entry` → implementation and independent Cursor review complete; explicit finalize remains separate.
+
 ## [2026-08-22.134737] - 2026-08-22
 
 ### Summary
diff --git a/README.md b/README.md
index f23f0ac..07769ad 100644
--- a/README.md
+++ b/README.md
@@ -138,7 +138,7 @@ Host support:
 | Claude | PROTOCOL_VERIFIED | `claude:en-US` / `claude:zh-CN` | Full capability continuation |
 | Qoder | PROTOCOL_VERIFIED | `qoder` | Validated on Qoder CLI |
 | Copilot | BASELINE_SUPPORTED | `copilot:en-US` / `copilot:zh-CN` | Prompt-only; payload uplift planned |
-| [Cursor](docs/cursor-host.md) | BASELINE_SUPPORTED | `cursor:en-US` / `cursor:zh-CN` | IDE user Plugin + global Skills/payload + user hooks |
+| [Cursor](docs/cursor-host.md) | BASELINE_SUPPORTED | `cursor:en-US` / `cursor:zh-CN` | IDE user Plugin + managed-first CLI entry + shared Skills/payload/hooks |
 
 Pass `--workspace <path>` to target another repo for workspace-scope or prewarm installs; Cursor installs once at user scope and does not require it. Use `--language <lang>` to control output language.
 
diff --git a/README.zh-CN.md b/README.zh-CN.md
index bba8fed..8b94efa 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -137,7 +137,7 @@ Get-Content sopify-install.ps1 | more
 | Claude | PROTOCOL_VERIFIED | `claude:zh-CN` / `claude:en-US` | 全能力接续 |
 | Qoder | PROTOCOL_VERIFIED | `qoder` | 已在 Qoder CLI 验证 |
 | Copilot | BASELINE_SUPPORTED | `copilot:zh-CN` / `copilot:en-US` | 仅 prompt；payload 升级计划中 |
-| [Cursor](docs/cursor-host.md) | BASELINE_SUPPORTED | `cursor:zh-CN` / `cursor:en-US` | IDE 用户 Plugin + 全局 Skills/payload + 用户 Hooks |
+| [Cursor](docs/cursor-host.md) | BASELINE_SUPPORTED | `cursor:zh-CN` / `cursor:en-US` | IDE 用户 Plugin + managed-first CLI 入口 + 共享 Skills/payload/Hooks |
 
 可用 `--workspace <path>` 指定目标仓库，`--language <lang>` 控制输出语言；Cursor 为用户级一次安装，不需要 `--workspace`。
 
diff --git a/docs/cursor-host.md b/docs/cursor-host.md
index 5cde51f..b099f25 100644
--- a/docs/cursor-host.md
+++ b/docs/cursor-host.md
@@ -1,6 +1,6 @@
 # Cursor 宿主接入
 
-主要支持本地 Cursor IDE。Cursor Agent CLI 不提供自动 Plugin Rule 入口，Cloud Agent 不在本期范围。支持档位保持 `BASELINE_SUPPORTED`。真实 IDE 已观察到 Plugin Rule 路由、Analyze 评分、AskQuestion 调用尝试与文本回退、managed develop、writer 写回和 Hook 拒绝明显直写；这些证据不代表跨模型稳定性或完整协议认证。本期不为 CLI 增加 launcher、项目 Rule 或 runtime。
+支持本地 Cursor IDE 与 Agent CLI，Cloud Agent 不在本期范围。IDE 通过用户级 Always Plugin Rule 进入；Agent CLI 默认安装 managed-first 顶层 Skill，被模型选中后实际读取同一 Rule。CLI 自动选择属于 best-effort，支持档位保持 `BASELINE_SUPPORTED`；本期不增加 launcher、项目 Rule 或 runtime，也不把安装存在外推为跨模型稳定性或完整协议认证。
 
 ## 安装面
 
@@ -13,7 +13,7 @@ curl -fsSL https://github.com/evidentloop/sopify/releases/latest/download/instal
 安装一次，管理四个用户级落点：
 
 - Plugin：`~/.cursor/plugins/local/sopify/.cursor-plugin/plugin.json`、`README.md` 与 `rules/sopify.mdc`。README 渐进介绍用途与边界；薄 Always Rule 是 IDE 的语义入口。
-- Skills：`~/.cursor/skills/sopify/` 下的 `analyze`、`design`、`develop`、`kb`、`templates` 和共享 references。
+- Skills：`~/.cursor/skills/sopify/SKILL.md` 是 Agent CLI 的 managed-first 顶层入口；同目录下的 `analyze`、`design`、`develop`、`kb`、`templates` 和共享 references 由 IDE 与 CLI 共用。
 - Payload：`~/.cursor/sopify/`，包含版本化 bundle 与 Hook helper。
 - Hooks：合并到 `~/.cursor/hooks.json`；不写仓库级 hooks。
 
@@ -23,10 +23,11 @@ Cursor 不需要 `--workspace`，安装器不在目标仓库创建 `.cursor/rule
 
 ## 语义与 Hook 边界
 
-IDE Plugin Rule 负责意图分类与 Skill 路由：
+IDE 直接加载 Plugin Rule；Agent CLI 的顶层 Skill 被选中后先读取同一 Rule。Rule 负责意图分类与 Skill 路由：
 
 - `consult_readonly` 直接只读回答，不自动恢复 active plan；
 - `quick_fix` 可修改用户授权的产品代码，但不直接写协议 state、handoff、receipt 或知识库；
+- 已授权动作若扩大为新的用户级或系统级持久化、删除操作，先说明新增影响并重新等待确认；
 - `new_plan`、`continue_plan`、`finalize` 和明确的 managed develop 才进入对应 Skill 与四步协议入口；
 - Analyze 必须实际执行 Skill 中的评分脚本，machine truth 只经 `sopify_writer` 库 API 写入。
 
@@ -47,7 +48,7 @@ python3 scripts/sopify_doctor.py --format json --home-root "$HOME"
 Cursor 分开报告：
 
 - `cursor_plugin_present`：Plugin manifest 与 Rule 是否存在；README 是包内说明，不作为行为或安装健康硬门槛；
-- `global_skill_tree_present`：五项 Skills、共享写作/输出契约与 Analyze 评分脚本是否存在；
+- `global_skill_tree_present`：CLI 顶层 Skill、五项阶段 Skills、共享写作/输出契约与 Analyze 评分脚本是否存在；
 - `payload_present`：payload/bundle 是否结构完整；
 - `cursor_hooks_present`：用户 hooks 与 helper 是否安全、完整；
 - `cursor_ide_behavior`、`cursor_cli_behavior`：真实宿主黑盒，默认 `skip (BLACK_BOX_NOT_VERIFIED)`。
@@ -60,10 +61,10 @@ IDE 与 CLI 分开记录，不互相外推：
 
 1. IDE baseline 已有可观察证据：已安装 Rule 含 `alwaysApply: true`，consult 实际读取 Cursor 写作规范并保持只读；Analyze 读取 Cursor Skill 并执行评分脚本；有限选项澄清留下 AskQuestion tool use 与 `Tool not found` 文本回退，结构化问卷 UI 仅作为人工观察，transcript 不含 tool result 或选项回传；managed 场景按 Analyze → Design → Develop 推进，经 `sopify_writer` 写 state、handoff 与 receipts；明显 Shell 直写 machine truth 被 Hook 拒绝且文件哈希不变。
 2. IDE 未验证边界：AskQuestion 是否跨模型稳定提供、独立 `sessionStart` 行为与显式 finalize。本期依赖文本追问回退和现有协议边界，不把这些项目作为 baseline 发布阻塞。
-3. CLI 未验证自动语义入口。CLI 的结果不能用于证明 IDE 行为，IDE 的结果也不能用于证明 CLI 行为。
+3. CLI 已观察到 managed 审计请求自动选择顶层 Skill、读取 Plugin Rule，并在修正 description 读取顺序后通过独立 Cursor 复审。该证据只证明当前本机路径，不承诺每个普通请求、所有模型或所有机器都会自动选择。
 
-Cursor 官方 CLI 文档明确项目 Rules，但没有把用户 Plugin Rule 写成 CLI 加载契约。2026-08-20 的真实会话也未读取 Cursor Plugin Rule 指定的共享规范，显式 `--plugin-dir` 唯一标记探针同样未生效。因此 CLI 自动入口明确不支持，也不阻塞本次 IDE baseline 发布。
+Cursor CLI 不会自行加载用户 Plugin Rule；Sopify 通过默认安装的顶层 Skill 显式读取该 Rule，不依赖 `--plugin-dir`、项目 Rule 或 prompt injector。Skill 发现和模型选择仍是宿主行为，因此 Doctor 继续把安装事实与 `cursor_cli_behavior` 黑盒证据分开。
 
-一次临时仓库黑盒已观察到四步读链、managed develop 与 `sopify_writer` 写回；显式 finalize 仍按用户命令触发，未纳入本次 baseline 黑盒。Doctor 的行为项继续保持静态 `BLACK_BOX_NOT_VERIFIED`，不把某台机器上的临时 transcript 写成可移植产品认证。
+一次临时仓库黑盒已观察到四步读链、managed develop 与 `sopify_writer` 写回；本机顶层 Skill 复审又观察到 managed 请求自动选择和 Rule 消费。显式 finalize、跨模型一致性与可移植黑盒仍未完整认证；Doctor 的行为项继续保持静态 `BLACK_BOX_NOT_VERIFIED`。
 
 官方依据：[`Cursor Plugins`](https://cursor.com/docs/plugins)、[`Plugin format`](https://cursor.com/docs/reference/plugins)、[`Cursor Rules`](https://cursor.com/docs/rules)、[`Cursor Agent Skills`](https://cursor.com/docs/skills)、[`Cursor Hooks`](https://cursor.com/docs/hooks)、[`Cursor CLI`](https://cursor.com/docs/cli/using)。
diff --git a/docs/getting-started.md b/docs/getting-started.md
index eac68a8..74b6eb7 100644
--- a/docs/getting-started.md
+++ b/docs/getting-started.md
@@ -139,11 +139,11 @@ Copilot reads project-level instruction files across its supported surfaces
 Full trigger wiring (equivalent to Codex/Claude `~go`) is coming in a future
 release.
 
-### Cursor IDE (optional Agent CLI)
+### Cursor IDE and local Agent CLI
 
 Cursor IDE uses one user-level Plugin rule. The same installation provides a
-global Skill/payload tree under `~/.cursor/` and user-level hooks in
-`~/.cursor/hooks.json`:
+managed-first Agent CLI entry, the shared Skill/payload tree under `~/.cursor/`,
+and user-level hooks in `~/.cursor/hooks.json`:
 
 ```bash
 curl -fsSL https://github.com/evidentloop/sopify/releases/latest/download/install.sh | bash -s -- --target cursor:en-US
@@ -157,11 +157,12 @@ created during install. `--with-evidentloop` is not supported for Cursor in
 this release. The Plugin Rule uses valid `.mdc` frontmatter with
 `alwaysApply: true` and routes to `~/.cursor/skills/sopify`.
 
-The tested Agent CLI does not automatically load this user Plugin Rule.
-Automatic entry for Agent CLI and Cloud Agent is outside this release. No
-launcher, project Rule, prompt injector, or runtime is added. See
-[Cursor host acceptance](./cursor-host.md). The current tier remains
-`BASELINE_SUPPORTED`.
+Agent CLI does not load the user Plugin Rule by itself. The installed top-level
+`~/.cursor/skills/sopify/SKILL.md` is a managed-first, model-selected entry that
+reads the Rule before routing phase Skills. Selection is best-effort; ordinary
+questions do not have to trigger it, and Cloud Agent remains outside this
+release. No launcher, project Rule, prompt injector, or runtime is added. See
+[Cursor host acceptance](./cursor-host.md). The tier remains `BASELINE_SUPPORTED`.
 
 ## Verify Setup
 
diff --git a/docs/how-sopify-works.en.md b/docs/how-sopify-works.en.md
index eb20d54..c7969b0 100644
--- a/docs/how-sopify-works.en.md
+++ b/docs/how-sopify-works.en.md
@@ -109,7 +109,7 @@ Layer notes:
 | Claude | PROTOCOL_VERIFIED | `install.sh --target claude:en-US` | Full capability continuation |
 | Qoder | PROTOCOL_VERIFIED | `install.sh --target qoder` | Validated on Qoder CLI |
 | Copilot | BASELINE_SUPPORTED | `install.sh --target copilot` | Prompt-only; payload uplift planned |
-| Cursor | BASELINE_SUPPORTED | `install.sh --target cursor:en-US` | IDE user Plugin + global Skills/payload + user hooks |
+| Cursor | BASELINE_SUPPORTED | `install.sh --target cursor:en-US` | IDE user Plugin + managed-first CLI entry + shared Skills/payload/hooks |
 
 ## Appendix: Plan Lifecycle
 
diff --git a/docs/how-sopify-works.md b/docs/how-sopify-works.md
index 337376d..e187b06 100644
--- a/docs/how-sopify-works.md
+++ b/docs/how-sopify-works.md
@@ -109,7 +109,7 @@ Sopify 不默认安装或运行验证器。只用 Sopify、使用其他 Verifier
 | Claude | PROTOCOL_VERIFIED | `install.sh --target claude:zh-CN` | 全能力接续 |
 | Qoder | PROTOCOL_VERIFIED | `install.sh --target qoder` | 已在 Qoder CLI 验证 |
 | Copilot | BASELINE_SUPPORTED | `install.sh --target copilot` | 仅 prompt；payload 升级计划中 |
-| Cursor | BASELINE_SUPPORTED | `install.sh --target cursor:zh-CN` | IDE 用户 Plugin + 全局 Skills/payload + 用户 Hooks |
+| Cursor | BASELINE_SUPPORTED | `install.sh --target cursor:zh-CN` | IDE 用户 Plugin + managed-first CLI 入口 + 共享 Skills/payload/Hooks |
 
 ## 附录：Plan 生命周期
 
diff --git a/index.html b/index.html
index a0a66d9..4ebb905 100644
--- a/index.html
+++ b/index.html
@@ -252,7 +252,7 @@ <h2>The short version</h2>
           </details>
           <details>
             <summary>Does switching hosts give every host identical capabilities?</summary>
-            <p>Codex, Claude, and Qoder have protocol-verified paths. Copilot and Cursor are baseline-supported. Cursor's automatic Plugin Rule entry applies to the local IDE, not Agent CLI or Cloud Agent.</p>
+            <p>Codex, Claude, and Qoder have protocol-verified paths. Copilot and Cursor are baseline-supported. One Cursor install provides the IDE Plugin Rule and a best-effort managed entry for local Agent CLI; Cloud Agent remains out of scope.</p>
           </details>
           <details>
             <summary>Is EvidentLoop required?</summary>
diff --git a/installer/cursor_plugin.py b/installer/cursor_plugin.py
index 97ba657..816b86e 100644
--- a/installer/cursor_plugin.py
+++ b/installer/cursor_plugin.py
@@ -11,6 +11,7 @@
 
 _IGNORE_PATTERNS = shutil.ignore_patterns(".DS_Store", "Thumbs.db", "__pycache__")
 _README_TEMPLATE_NAME = "cursor-plugin-readme.md.template"
+_CLI_SKILL_TEMPLATE_NAME = "cursor-cli-skill.md.template"
 
 
 def install_cursor_user_plugin_assets(
@@ -26,11 +27,14 @@ def install_cursor_user_plugin_assets(
 
     rule_source = adapter.instruction_source(repo_root, language_directory)
     readme_source = rule_source.with_name(_README_TEMPLATE_NAME)
+    root_skill_source = rule_source.with_name(_CLI_SKILL_TEMPLATE_NAME)
     skills_source = adapter.source_root(repo_root, language_directory) / "skills" / "sopify"
     if not rule_source.is_file():
         raise InstallError(f"Missing source Cursor Plugin rule: {rule_source}")
     if not readme_source.is_file():
         raise InstallError(f"Missing source Cursor Plugin README: {readme_source}")
+    if not root_skill_source.is_file():
+        raise InstallError(f"Missing source Cursor CLI Skill: {root_skill_source}")
     if not skills_source.is_dir():
         raise InstallError(f"Missing source skills directory: {skills_source}")
 
@@ -38,9 +42,11 @@ def install_cursor_user_plugin_assets(
     readme_path = adapter.user_plugin_readme_path(home_root)
     plugin_root = rule_path.parent.parent
     skills_destination = adapter.destination_root(home_root) / "skills" / "sopify"
+    root_skill_destination = skills_destination / "SKILL.md"
     manifest = _manifest_text()
     rule = render_user_plugin_rule(rule_source, adapter)
     readme = readme_source.read_text(encoding="utf-8").rstrip("\n") + "\n"
+    root_skill = root_skill_source.read_text(encoding="utf-8").rstrip("\n") + "\n"
     expected_paths = adapter.expected_paths(home_root)
     if (
         manifest_path.is_file()
@@ -49,6 +55,8 @@ def install_cursor_user_plugin_assets(
         and rule_path.read_text(encoding="utf-8") == rule
         and readme_path.is_file()
         and readme_path.read_text(encoding="utf-8") == readme
+        and root_skill_destination.is_file()
+        and root_skill_destination.read_text(encoding="utf-8") == root_skill
         and all(path.exists() for path in adapter.global_skill_paths(home_root))
     ):
         return InstallPhaseResult(
@@ -71,6 +79,7 @@ def install_cursor_user_plugin_assets(
         shutil.rmtree(skills_destination)
     skills_destination.parent.mkdir(parents=True, exist_ok=True)
     shutil.copytree(skills_source, skills_destination, ignore=_IGNORE_PATTERNS)
+    root_skill_destination.write_text(root_skill, encoding="utf-8")
 
     return InstallPhaseResult(
         action=action,
diff --git a/installer/hosts/base.py b/installer/hosts/base.py
index 44dcc0c..dec2b62 100644
--- a/installer/hosts/base.py
+++ b/installer/hosts/base.py
@@ -91,6 +91,7 @@ def workspace_expected_paths(self, workspace_root: Path) -> tuple[Path, ...]:
     def global_skill_paths(self, home_root: Path) -> tuple[Path, ...]:
         skills_root = self.destination_root(home_root) / "skills" / "sopify"
         return (
+            skills_root / "SKILL.md",
             skills_root / "analyze" / "SKILL.md",
             skills_root / "design" / "SKILL.md",
             skills_root / "develop" / "SKILL.md",
diff --git a/installer/inspection.py b/installer/inspection.py
index 195f406..123c289 100644
--- a/installer/inspection.py
+++ b/installer/inspection.py
@@ -840,8 +840,8 @@ def _cursor_behavior_check(*, capability: HostCapability, check_id: str) -> Insp
         )
     else:
         recommendation = (
-            "Optionally run the Cursor Agent CLI manual-Skill and Hook black-box flow. "
-            "Automatic loading of the user Plugin Rule is not a supported CLI entry."
+            "Run the Cursor Agent CLI managed-entry and Hook black-box flow. "
+            "Top-level Skill selection is best-effort and remains separate from file installation evidence."
         )
     return _cursor_not_verified_check(
         capability=capability,
diff --git a/skills/en/cursor-cli-skill.md.template b/skills/en/cursor-cli-skill.md.template
new file mode 100644
index 0000000..5c022db
--- /dev/null
+++ b/skills/en/cursor-cli-skill.md.template
@@ -0,0 +1,23 @@
+---
+name: sopify
+description: Top-level entry for Sopify managed workflows in Cursor Agent CLI. Use it for .sopify, ~go, requirements analysis, solution design, development, plan continuation or resumption, and finalize or closeout requests; once selected, first load only this Skill and the installed Plugin Rule without parallel business reads or other tools. Ordinary software questions and unmanaged quick fixes do not require it.
+---
+
+# Sopify CLI Top-Level Entry
+
+This Skill adds a managed-first entry for Cursor Agent CLI. It only loads the installed Sopify protocol and routes phase Skills; it does not copy the protocol or replace a Plugin Rule already loaded by Cursor IDE.
+
+## Entry steps
+
+1. If the current context already contains the `Sopify Cursor IDE Entry` heading and the five intents `consult_readonly`, `quick_fix`, `new_plan`, `continue_plan`, and `finalize`, follow that protocol without reading it again.
+2. Otherwise, after loading this Skill, read only `~/.cursor/plugins/local/sopify/rules/sopify.mdc` next. Do not read business files or call other tools in parallel. Read the file for real rather than merely claiming to reference it; treat its frontmatter as Cursor metadata and its body as the top-level Sopify protocol for this conversation.
+3. Classify the current request using that protocol, then read the relevant phase Skill, knowledge-base guidance, shared writing guidance, and output contract under `~/.cursor/skills/sopify/` only as needed.
+4. If the Plugin Rule is missing or unreadable, stop Sopify routing and ask the user to rerun the Cursor installer for the current language. Do not substitute a same-named Skill from `~/.claude`, `~/.codex`, or the project.
+
+## Boundaries
+
+- Automatic selection is model-driven and best-effort. Claim that Sopify was entered only after this Skill or the Plugin Rule was actually read.
+- Ordinary software questions and unmanaged quick fixes do not have to trigger this Skill. Never pretend that an unmatched managed request consumed Sopify.
+- Intent classification, scoring, the four-step protocol, knowledge-base policy, output, and writer boundaries remain authoritative in the installed Plugin Rule and phase Skills; this file does not duplicate them.
+- If an authorized action expands into new user-level or system-level persistence or deletion, explain the added impact and wait for confirmation again.
+- Do not modify the Cursor Plugin Rule, Hooks, settings, CLI configuration, proxy, model, account, API key, keychain, or MCP.
diff --git a/skills/en/cursor-plugin-readme.md.template b/skills/en/cursor-plugin-readme.md.template
index 4c52abd..cb045af 100644
--- a/skills/en/cursor-plugin-readme.md.template
+++ b/skills/en/cursor-plugin-readme.md.template
@@ -1,6 +1,6 @@
 # Sopify
 
-Sopify is an adaptive workflow layer for Cursor IDE. It routes each request to the smallest appropriate path—read-only consultation, quick fix, analysis, design, or managed development—while keeping evidence and project-state boundaries explicit.
+Sopify is an adaptive workflow layer for Cursor IDE and local Agent CLI. It routes each request to the smallest appropriate path—read-only consultation, quick fix, analysis, design, or managed development—while keeping evidence and project-state boundaries explicit.
 
 ## Use Sopify
 
@@ -8,12 +8,13 @@ Describe the goal naturally. Sopify uses the current repository's instructions a
 
 ## How it works
 
-- **Always Rule** classifies intent and routes the request.
+- **IDE Always Rule** classifies intent and routes requests in Cursor IDE.
+- **CLI top-level Skill** reads the same Rule for managed requests in local Agent CLI; model selection is best-effort.
 - **Skills** provide the phase-specific procedures.
 - **User Hooks** guard protocol state and receipt files against obvious direct writes; they are not a security sandbox.
 
 ## Support boundary
 
-- Cursor IDE is the automatic semantic entry. Cursor Agent CLI uses the installed Skills manually.
-- Support remains `BASELINE_SUPPORTED`: the primary IDE path has observable evidence, but automatic CLI entry, Cloud Agents, and identical behavior across models are not promised.
+- Cursor IDE uses the Always Rule. Cursor Agent CLI installs a managed-first top-level Skill by default and reuses the same phase Skills, payload, and user Hooks.
+- Support remains `BASELINE_SUPPORTED`: IDE and CLI evidence stay separate, and Sopify does not promise that every ordinary request selects the top-level Skill, Cloud Agents, or identical behavior across models.
 - Sopify does not modify Cursor settings, proxy, model, account, API keys, keychain, or MCP configuration. Cloud Agents are out of scope.
diff --git a/skills/en/cursor-plugin-rule.mdc.template b/skills/en/cursor-plugin-rule.mdc.template
index e7d4b14..20b20d0 100644
--- a/skills/en/cursor-plugin-rule.mdc.template
+++ b/skills/en/cursor-plugin-rule.mdc.template
@@ -6,7 +6,7 @@ alwaysApply: true
 
 # Sopify Cursor IDE Entry
 
-This is the user-level Cursor Plugin rule for the IDE. When the IDE loads it, it is the Sopify semantic entry. It governs intent routing, protocol boundaries, and output; the installed Sopify Skills retain the phase procedures. The Agent CLI does not automatically load this Rule in the validated release and is outside this automatic-entry contract.
+This is the user-level Cursor Plugin rule for the IDE. When the IDE loads it, it is the Sopify semantic entry. It governs intent routing, protocol boundaries, and output; the installed Sopify Skills retain the phase procedures. Agent CLI does not load a Plugin Rule by itself; when the default top-level `sopify` Skill is selected, it reads this Rule and reuses the same protocol.
 
 ## Entry for every Agent request
 
@@ -14,7 +14,8 @@ This is the user-level Cursor Plugin rule for the IDE. When the IDE loads it, it
 2. Classify intent first: `consult_readonly`, `quick_fix`, `new_plan`, `continue_plan`, or `finalize`. Never resume merely because an active plan exists.
 3. Treat `AGENTS.md`, `CLAUDE.md`, project Rules, and `.sopify/` in the current target repository as project facts. A multi-repository parent directory is not automatically the target repository.
 4. Answer `consult_readonly` directly and read-only without loading a phase Skill. `quick_fix` may modify user-authorized product code, but neither mode resumes an old plan or directly writes state, handoff, receipts, or the knowledge base.
-5. For workflow work, read and execute the exact `~/.cursor` path listed below. Do not substitute a same-named Skill from `~/.claude`, `~/.codex`, or another compatibility directory:
+5. If an authorized action expands into new user-level or system-level persistence or deletion, explain the added impact and wait for confirmation again.
+6. For workflow work, read and execute the exact `~/.cursor` path listed below. Do not substitute a same-named Skill from `~/.claude`, `~/.codex`, or another compatibility directory:
    - Requirements analysis: `{{skills_root}}/analyze/SKILL.md`
    - Solution design: `{{skills_root}}/design/SKILL.md`
    - Development: `{{skills_root}}/develop/SKILL.md`
@@ -32,6 +33,6 @@ This is the user-level Cursor Plugin rule for the IDE. When the IDE loads it, it
 ## Support boundary and output
 
 - Cursor remains `BASELINE_SUPPORTED`: file presence or a single model claim is not protocol-level verification; support conclusions require observable host evidence.
-- Agent CLI users may invoke the installed Skills manually; user Hooks remain available where Cursor loads them. Sopify does not claim automatic CLI routing through this Plugin Rule.
+- Agent CLI installs a managed-first top-level Skill by default. Model selection is best-effort, and Sopify is entered only after the top-level Skill and this Rule were actually read. User Hooks remain available where Cursor loads them.
 - Follow the already-read `{{skills_root}}/references/shared-writing-dna.md`. After entering a phase Skill, read `{{skills_root}}/references/output-contract.md`. For consult, give only the conclusion and necessary evidence; do not add a banner, `Changes`, or `Next` to fill a template.
 - Do not modify Cursor settings, proxy, model, account, API key, keychain, or MCP. Cloud Agents are out of scope.
diff --git a/skills/zh/cursor-cli-skill.md.template b/skills/zh/cursor-cli-skill.md.template
new file mode 100644
index 0000000..ad538a7
--- /dev/null
+++ b/skills/zh/cursor-cli-skill.md.template
@@ -0,0 +1,23 @@
+---
+name: sopify
+description: Cursor Agent CLI 中 Sopify 托管工作流的顶层入口。请求涉及 .sopify、~go、需求分析、方案设计、开发实施、继续、接续、恢复方案或 finalize、收口时优先使用；命中后第一步只加载本 Skill 和已安装 Plugin Rule，不与业务读取或其他工具并行。普通软件咨询和非托管 quick fix 不强制触发。
+---
+
+# Sopify CLI 顶层入口
+
+此 Skill 为 Cursor Agent CLI 补充 managed-first 入口。它只负责加载当前安装的 Sopify 协议并路由阶段 Skill，不复制协议正文，也不替换 Cursor IDE 已加载的 Plugin Rule。
+
+## 入口步骤
+
+1. 如果当前上下文已包含标题 `Sopify Cursor IDE 入口`，以及 `consult_readonly`、`quick_fix`、`new_plan`、`continue_plan`、`finalize` 五类意图，直接遵循该协议，不重复读取。
+2. 否则，加载本 Skill 后的下一步只读取 `~/.cursor/plugins/local/sopify/rules/sopify.mdc`；不要并行读取业务文件或调用其他工具。必须实际读取，不得只声称已参考；frontmatter 是 Cursor 元数据，正文是当前会话的顶层 Sopify 协议。
+3. 先按该协议分类本轮意图，再按需读取 `~/.cursor/skills/sopify/` 下对应的阶段 Skill、KB、共享写作规范与输出契约。
+4. Plugin Rule 缺失或不可读时停止 Sopify 路由，提示用户重新运行当前语言的 Cursor 安装；不得用 `~/.claude`、`~/.codex` 或项目中的同名 Skill 顶替。
+
+## 边界
+
+- 自动选择由 Cursor 模型决定，属于 best-effort。只有实际读取本 Skill 或 Plugin Rule 后，才可以声称本轮已进入 Sopify。
+- 普通软件咨询和非托管 quick fix 不强制触发；managed 请求未命中时，不得伪装成已消费 Sopify。
+- 意图分类、评分门、四步协议、KB、输出和 writer 边界均以已安装 Plugin Rule 与阶段 Skills 为准，本文件不重复定义。
+- 如果已授权动作扩大为新的用户级或系统级持久化、删除操作，先说明新增影响并重新等待确认。
+- 不修改 Cursor Plugin Rule、Hooks、settings、CLI config、代理、模型、账号、API Key、钥匙串或 MCP。
diff --git a/skills/zh/cursor-plugin-readme.md.template b/skills/zh/cursor-plugin-readme.md.template
index 50b974f..2d8a024 100644
--- a/skills/zh/cursor-plugin-readme.md.template
+++ b/skills/zh/cursor-plugin-readme.md.template
@@ -1,6 +1,6 @@
 # Sopify
 
-Sopify 是 Cursor IDE 的自适应工作流层。它先把请求路由到最小适用路径——只读咨询、快速修复、需求分析、方案设计或受管开发——并明确约束证据和项目状态边界。
+Sopify 是 Cursor IDE 与本地 Agent CLI 的自适应工作流层。它先把请求路由到最小适用路径——只读咨询、快速修复、需求分析、方案设计或受管开发——并明确约束证据和项目状态边界。
 
 ## 使用方式
 
@@ -8,12 +8,13 @@ Sopify 是 Cursor IDE 的自适应工作流层。它先把请求路由到最小
 
 ## 工作方式
 
-- **Always Rule**：分类意图并路由请求。
+- **IDE Always Rule**：在 Cursor IDE 中分类意图并路由请求。
+- **CLI 顶层 Skill**：在本地 Agent CLI 中为 managed 请求读取同一 Rule；模型选择属于 best-effort。
 - **Skills**：提供各阶段的执行细则。
 - **用户级 Hooks**：对协议 state、handoff 与 receipt 提供明显直写保护；它们不是安全沙箱。
 
 ## 支持边界
 
-- Cursor IDE 是自动语义入口；Cursor Agent CLI 手工使用已安装 Skills。
-- 当前支持档位为 `BASELINE_SUPPORTED`；IDE 主路径已有可观察证据，但不承诺 CLI 自动入口、Cloud Agent 或跨模型完全一致。
+- Cursor IDE 使用 Always Rule；Cursor Agent CLI 默认安装 managed-first 顶层 Skill，并复用同一阶段 Skills、payload 与用户 Hooks。
+- 当前支持档位为 `BASELINE_SUPPORTED`；IDE 与 CLI 证据分开记录，不承诺每个普通请求都选择顶层 Skill、Cloud Agent 或跨模型完全一致。
 - Sopify 不修改 Cursor settings、代理、模型、账号、API Key、钥匙串或 MCP 配置；不支持 Cloud Agent。
diff --git a/skills/zh/cursor-plugin-rule.mdc.template b/skills/zh/cursor-plugin-rule.mdc.template
index 151e15b..265aea5 100644
--- a/skills/zh/cursor-plugin-rule.mdc.template
+++ b/skills/zh/cursor-plugin-rule.mdc.template
@@ -6,7 +6,7 @@ alwaysApply: true
 
 # Sopify Cursor IDE 入口
 
-这是 Cursor IDE 的用户级 Plugin Rule；IDE 实际加载它时，它作为 Sopify 语义入口。它只负责意图路由、协议边界和输出治理；阶段细则仍由已安装的 Sopify Skills 提供。已验证版本的 Agent CLI 不会自动加载此 Rule，因此不属于本自动入口契约。
+这是 Cursor IDE 的用户级 Plugin Rule；IDE 实际加载它时，它作为 Sopify 语义入口。它只负责意图路由、协议边界和输出治理；阶段细则仍由已安装的 Sopify Skills 提供。Agent CLI 不会自行加载 Plugin Rule；默认安装的顶层 `sopify` Skill 被模型选中后，会实际读取本 Rule 并复用同一协议。
 
 ## 每次 Agent 请求的入口
 
@@ -14,7 +14,8 @@ alwaysApply: true
 2. 先分类意图：`consult_readonly`、`quick_fix`、`new_plan`、`continue_plan` 或 `finalize`。不得因存在 active plan 就自动恢复。
 3. 当前目标仓库中的 `AGENTS.md`、`CLAUDE.md`、项目 Rules 和 `.sopify/` 都是项目事实；多仓父目录不自动等同于目标仓库。
 4. `consult_readonly` 直接只读回答，不加载阶段 Skill；`quick_fix` 可以修改用户授权的产品代码，但两者都不自动接续旧方案，也不直接写 state、handoff、receipt 或知识库。
-5. 需要进入工作流时，读取并执行下面列出的 `~/.cursor` 精确路径；不得替换为 `~/.claude`、`~/.codex` 或其他兼容目录中的同名 Skill：
+5. 如果已授权动作扩大为新的用户级或系统级持久化、删除操作，先说明新增影响并重新等待确认。
+6. 需要进入工作流时，读取并执行下面列出的 `~/.cursor` 精确路径；不得替换为 `~/.claude`、`~/.codex` 或其他兼容目录中的同名 Skill：
    - 需求分析：`{{skills_root}}/analyze/SKILL.md`
    - 方案设计：`{{skills_root}}/design/SKILL.md`
    - 开发实施：`{{skills_root}}/develop/SKILL.md`
@@ -32,6 +33,6 @@ alwaysApply: true
 ## 支持边界与输出
 
 - Cursor 首发保持 `BASELINE_SUPPORTED`：文件存在或单次模型自述不等于协议级验证；支持结论以可观察宿主证据为准。
-- Agent CLI 可按需手工调用已安装的 Skills；Cursor 加载用户 Hooks 时继续保留写保护。Sopify 不宣称此 Plugin Rule 能为 CLI 提供自动路由。
+- Agent CLI 默认安装 managed-first 顶层 Skill；模型自动选择属于 best-effort，只有实际读取顶层 Skill 与本 Rule 后才算进入 Sopify。Cursor 加载用户 Hooks 时继续保留写保护。
 - 表达遵循已读的 `{{skills_root}}/references/shared-writing-dna.md`。进入阶段 Skill 后再读 `{{skills_root}}/references/output-contract.md`；咨询只给结论和必要证据，不要为凑模板添加标题、`Changes` 或 `Next`。
 - 不修改 Cursor settings、代理、模型、账号、API Key、钥匙串或 MCP；不启用 Cloud Agent。
diff --git a/tests/test_installer_status_doctor.py b/tests/test_installer_status_doctor.py
index f68d536..3f859e1 100644
--- a/tests/test_installer_status_doctor.py
+++ b/tests/test_installer_status_doctor.py
@@ -19,6 +19,7 @@
 from installer.hosts.claude import CLAUDE_ADAPTER
 from installer.hosts.codex import CODEX_ADAPTER
 from installer.hosts.cursor import CURSOR_ADAPTER
+from installer.hosts.qoder import QODER_ADAPTER
 from installer.inspection import build_doctor_payload, build_status_payload, render_doctor_text, render_status_text
 from installer.models import InstallError
 from installer.payload import _REQUIRED_BUNDLE_CAPABILITIES, install_global_payload, run_workspace_bootstrap
@@ -135,6 +136,7 @@ def test_cursor_install_proves_split_paths_without_touching_settings(self) -> No
             self.assertNotIn("rules", manifest_payload)
             self.assertIn("adaptive workflow layer", readme.read_text(encoding="utf-8"))
             self.assertIn("BASELINE_SUPPORTED", readme.read_text(encoding="utf-8"))
+            self.assertTrue((skill_root / "SKILL.md").is_file())
             self.assertTrue((skill_root / "analyze" / "SKILL.md").is_file())
             self.assertTrue((skill_root / "design" / "SKILL.md").is_file())
             self.assertTrue((skill_root / "develop" / "SKILL.md").is_file())
@@ -156,7 +158,7 @@ def test_cursor_install_proves_split_paths_without_touching_settings(self) -> No
             self.assertIn("alwaysApply: true", rule.read_text(encoding="utf-8"))
             self.assertIn("~/.cursor/skills/sopify/references/shared-writing-dna.md", rule.read_text(encoding="utf-8"))
             self.assertIn("Cursor IDE", rule.read_text(encoding="utf-8"))
-            self.assertIn("does not automatically load", rule.read_text(encoding="utf-8"))
+            self.assertIn("does not load a Plugin Rule by itself", rule.read_text(encoding="utf-8"))
             self.assertNotIn(str(skill_root), rule.read_text(encoding="utf-8"))
             self.assertIsNone(result.workspace_bootstrap)
             self.assertEqual(result.target.host, "cursor")
@@ -216,6 +218,59 @@ def test_cursor_install_proves_split_paths_without_touching_settings(self) -> No
             self.assertEqual(repair_result.host_install.action, "updated")
             self.assertTrue(readme.is_file())
 
+    def test_cursor_reinstall_replaces_stale_cli_entry_and_cleans_backup(self) -> None:
+        with tempfile.TemporaryDirectory() as home_dir:
+            home_root = Path(home_dir)
+            run_install(
+                target_value="cursor:zh-CN",
+                workspace_value=None,
+                repo_root=REPO_ROOT,
+                home_root=home_root,
+            )
+
+            skill_root = home_root / ".cursor" / "skills" / "sopify"
+            root_skill = skill_root / "SKILL.md"
+            stale_backup = skill_root / "SKILL.md.old"
+            root_skill.write_text("old local probe\n", encoding="utf-8")
+            stale_backup.write_text("stale backup\n", encoding="utf-8")
+
+            updated = run_install(
+                target_value="cursor:zh-CN",
+                workspace_value=None,
+                repo_root=REPO_ROOT,
+                home_root=home_root,
+            )
+            canonical = (
+                REPO_ROOT / "skills" / "zh" / "cursor-cli-skill.md.template"
+            ).read_text(encoding="utf-8")
+
+            self.assertEqual(updated.host_install.action, "updated")
+            self.assertEqual(root_skill.read_text(encoding="utf-8"), canonical)
+            self.assertFalse(stale_backup.exists())
+
+            repeated = run_install(
+                target_value="cursor:zh-CN",
+                workspace_value=None,
+                repo_root=REPO_ROOT,
+                home_root=home_root,
+            )
+            self.assertEqual(repeated.host_install.action, "skipped")
+
+    def test_cursor_cli_entry_is_not_installed_for_other_home_hosts(self) -> None:
+        for adapter in (CODEX_ADAPTER, CLAUDE_ADAPTER, QODER_ADAPTER):
+            with self.subTest(host=adapter.host_name), tempfile.TemporaryDirectory() as home_dir:
+                home_root = Path(home_dir)
+                install_host_assets(
+                    adapter,
+                    repo_root=REPO_ROOT,
+                    home_root=home_root,
+                    language_directory="CN",
+                )
+                cursor_only_entry = (
+                    adapter.destination_root(home_root) / "skills" / "sopify" / "SKILL.md"
+                )
+                self.assertFalse(cursor_only_entry.exists())
+
     def test_cursor_install_preflights_invalid_hooks_before_any_product_write(self) -> None:
         with tempfile.TemporaryDirectory() as home_dir, tempfile.TemporaryDirectory() as workspace_dir:
             home_root = Path(home_dir)
diff --git a/tests/test_plan_contract_assets.py b/tests/test_plan_contract_assets.py
index 617de36..0eadf36 100644
--- a/tests/test_plan_contract_assets.py
+++ b/tests/test_plan_contract_assets.py
@@ -160,7 +160,9 @@ def test_cursor_plugin_rule_uses_legal_frontmatter_and_thin_routing_contract(sel
                 self.assertIn("consult_readonly", rendered)
                 self.assertIn("IDE Entry" if language == "en" else "IDE 入口", rendered)
                 self.assertIn(
-                    "does not automatically load" if language == "en" else "不会自动加载",
+                    "does not load a Plugin Rule by itself"
+                    if language == "en"
+                    else "不会自行加载 Plugin Rule",
                     rendered,
                 )
                 self.assertIn("~/.claude", rendered)
diff --git a/zh-CN.html b/zh-CN.html
index 000668e..9cedca9 100644
--- a/zh-CN.html
+++ b/zh-CN.html
@@ -245,7 +245,7 @@ <h2>先说最重要的</h2>
           </details>
           <details>
             <summary>换宿主后能力完全一样吗？</summary>
-            <p>Codex、Claude 和 Qoder 已通过协议路径验证。Copilot 和 Cursor 为基线支持；Cursor 的 Plugin Rule 自动入口只适用于本地 IDE，不覆盖 Agent CLI 或 Cloud Agent。</p>
+            <p>Codex、Claude 和 Qoder 已通过协议路径验证。Copilot 和 Cursor 为基线支持；一次 Cursor 安装同时提供 IDE Plugin Rule 与本地 Agent CLI 的 best-effort managed 入口，Cloud Agent 仍不在范围内。</p>
           </details>
           <details>
             <summary>必须安装 EvidentLoop 吗？</summary>

From bb645d1441b681f21bbd0ff1a0a1af7e6307b1b9 Mon Sep 17 00:00:00 2001
From: "sanze.li" <weixin.li@nio.com>
Date: Sat, 29 Aug 2026 23:46:46 +0800
Subject: [PATCH 2/3] docs: clarify Cursor IDE and CLI support

---
 .sopify/blueprint/README.md                |  2 +-
 CHANGELOG.md                               |  4 ++--
 README.md                                  |  2 +-
 README.zh-CN.md                            |  2 +-
 docs/cursor-host.md                        |  8 ++++----
 docs/getting-started.md                    | 20 +++++++++++---------
 docs/how-sopify-works.en.md                |  2 +-
 docs/how-sopify-works.md                   |  2 +-
 index.html                                 |  2 +-
 skills/en/cursor-plugin-readme.md.template | 20 ++++++++++----------
 skills/zh/cursor-plugin-readme.md.template | 20 ++++++++++----------
 tests/test_installer_status_doctor.py      |  5 +++--
 zh-CN.html                                 |  2 +-
 13 files changed, 47 insertions(+), 44 deletions(-)

diff --git a/.sopify/blueprint/README.md b/.sopify/blueprint/README.md
index 0fe07a5..7189453 100644
--- a/.sopify/blueprint/README.md
+++ b/.sopify/blueprint/README.md
@@ -13,7 +13,7 @@
 ## 当前焦点
 
 <!-- sopify:auto:focus:start -->
-- 当前活动 plan：`../plan/20260828_cursor_agent_cli_entry` 已完成 Cursor Agent CLI managed-first 顶层入口与一次安装产品化，本机安装和两轮 Cursor 独立审计通过，状态为 `ready_to_archive`；feat 分支 commit/push 已获授权，尚未 finalize、合并或发布。
+- 当前活动 plan：`../plan/20260828_cursor_agent_cli_entry` 已完成 Cursor IDE + 本地 CLI 一次安装，以及 CLI 方案/开发/接续入口；本机安装和两轮 Cursor 独立审计通过，状态为 `ready_to_archive`。feat 分支已提交并 push，PR #73 已创建；尚未 finalize、合并或发布。
 - MCP tool plane 已完成 Codex-first 持久注册验证和 Claude 会话级只读调用验证，Qoder / Claude / Copilot 的后续持久注册仍按证据推进。
 - P8 Protocol Kernel & Runtime Retirement 已归档至 `../history/2026-06/20260605_p8_protocol_kernel_runtime_retirement`。
 - history 归档：已可用；最近归档：`../history/2026-07/20260720_plain_language_output`。
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 7db1764..e86edb5 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,11 +8,11 @@ Format: Summary → Changed → Plan Packages. File-level details live in `git l
 
 ### Summary
 
-Extended the existing Cursor user-scope installation with a managed-first entry for local Agent CLI.
+Extended the existing Cursor user-scope installation so one install covers the IDE and local Agent CLI.
 
 ### Changed
 
-- **Cursor Agent CLI**: Added bilingual top-level Skills that read the installed Plugin Rule for managed requests, kept automatic model selection best-effort, and preserved separate IDE/CLI behavior evidence.
+- **Cursor Agent CLI**: Added bilingual top-level Skills for plan, development, and resume requests. The CLI selects the entry when relevant, reads the installed Plugin Rule, and keeps IDE/CLI behavior evidence separate.
 - **Cursor installation**: The existing `cursor:<language>` target now installs and verifies the CLI entry, replaces stale local copies on reinstall, and keeps settings, proxy, model, account, and Cloud Agent out of scope.
 
 ### Plan Packages
diff --git a/README.md b/README.md
index 07769ad..1589833 100644
--- a/README.md
+++ b/README.md
@@ -138,7 +138,7 @@ Host support:
 | Claude | PROTOCOL_VERIFIED | `claude:en-US` / `claude:zh-CN` | Full capability continuation |
 | Qoder | PROTOCOL_VERIFIED | `qoder` | Validated on Qoder CLI |
 | Copilot | BASELINE_SUPPORTED | `copilot:en-US` / `copilot:zh-CN` | Prompt-only; payload uplift planned |
-| [Cursor](docs/cursor-host.md) | BASELINE_SUPPORTED | `cursor:en-US` / `cursor:zh-CN` | IDE user Plugin + managed-first CLI entry + shared Skills/payload/hooks |
+| [Cursor](docs/cursor-host.md) | BASELINE_SUPPORTED | `cursor:en-US` / `cursor:zh-CN` | One install for IDE and local CLI; plan/development/resume/closeout entry is model-selected |
 
 Pass `--workspace <path>` to target another repo for workspace-scope or prewarm installs; Cursor installs once at user scope and does not require it. Use `--language <lang>` to control output language.
 
diff --git a/README.zh-CN.md b/README.zh-CN.md
index 8b94efa..80afdf7 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -137,7 +137,7 @@ Get-Content sopify-install.ps1 | more
 | Claude | PROTOCOL_VERIFIED | `claude:zh-CN` / `claude:en-US` | 全能力接续 |
 | Qoder | PROTOCOL_VERIFIED | `qoder` | 已在 Qoder CLI 验证 |
 | Copilot | BASELINE_SUPPORTED | `copilot:zh-CN` / `copilot:en-US` | 仅 prompt；payload 升级计划中 |
-| [Cursor](docs/cursor-host.md) | BASELINE_SUPPORTED | `cursor:zh-CN` / `cursor:en-US` | IDE 用户 Plugin + managed-first CLI 入口 + 共享 Skills/payload/Hooks |
+| [Cursor](docs/cursor-host.md) | BASELINE_SUPPORTED | `cursor:zh-CN` / `cursor:en-US` | 一次安装支持 IDE 与本地 CLI；CLI 方案/开发/接续/收口入口按需触发 |
 
 可用 `--workspace <path>` 指定目标仓库，`--language <lang>` 控制输出语言；Cursor 为用户级一次安装，不需要 `--workspace`。
 
diff --git a/docs/cursor-host.md b/docs/cursor-host.md
index b099f25..dcdcd33 100644
--- a/docs/cursor-host.md
+++ b/docs/cursor-host.md
@@ -1,6 +1,6 @@
 # Cursor 宿主接入
 
-支持本地 Cursor IDE 与 Agent CLI，Cloud Agent 不在本期范围。IDE 通过用户级 Always Plugin Rule 进入；Agent CLI 默认安装 managed-first 顶层 Skill，被模型选中后实际读取同一 Rule。CLI 自动选择属于 best-effort，支持档位保持 `BASELINE_SUPPORTED`；本期不增加 launcher、项目 Rule 或 runtime，也不把安装存在外推为跨模型稳定性或完整协议认证。
+Cursor 安装一次即可支持本地 IDE 与 Agent CLI，Cloud Agent 不在本期范围。IDE 始终通过用户级 Plugin Rule 进入；CLI 在方案、开发、继续和收口等 Sopify 请求中按需选择顶层 Skill，再读取同一 Rule。普通问答不强制经过该入口。支持档位保持 `BASELINE_SUPPORTED`；本期不增加 launcher、项目 Rule 或 runtime，也不把安装成功外推为跨模型稳定性或完整协议认证。
 
 ## 安装面
 
@@ -13,7 +13,7 @@ curl -fsSL https://github.com/evidentloop/sopify/releases/latest/download/instal
 安装一次，管理四个用户级落点：
 
 - Plugin：`~/.cursor/plugins/local/sopify/.cursor-plugin/plugin.json`、`README.md` 与 `rules/sopify.mdc`。README 渐进介绍用途与边界；薄 Always Rule 是 IDE 的语义入口。
-- Skills：`~/.cursor/skills/sopify/SKILL.md` 是 Agent CLI 的 managed-first 顶层入口；同目录下的 `analyze`、`design`、`develop`、`kb`、`templates` 和共享 references 由 IDE 与 CLI 共用。
+- Skills：`~/.cursor/skills/sopify/SKILL.md` 是 Agent CLI 的方案/开发/接续入口；同目录下的 `analyze`、`design`、`develop`、`kb`、`templates` 和共享 references 由 IDE 与 CLI 共用。
 - Payload：`~/.cursor/sopify/`，包含版本化 bundle 与 Hook helper。
 - Hooks：合并到 `~/.cursor/hooks.json`；不写仓库级 hooks。
 
@@ -61,10 +61,10 @@ IDE 与 CLI 分开记录，不互相外推：
 
 1. IDE baseline 已有可观察证据：已安装 Rule 含 `alwaysApply: true`，consult 实际读取 Cursor 写作规范并保持只读；Analyze 读取 Cursor Skill 并执行评分脚本；有限选项澄清留下 AskQuestion tool use 与 `Tool not found` 文本回退，结构化问卷 UI 仅作为人工观察，transcript 不含 tool result 或选项回传；managed 场景按 Analyze → Design → Develop 推进，经 `sopify_writer` 写 state、handoff 与 receipts；明显 Shell 直写 machine truth 被 Hook 拒绝且文件哈希不变。
 2. IDE 未验证边界：AskQuestion 是否跨模型稳定提供、独立 `sessionStart` 行为与显式 finalize。本期依赖文本追问回退和现有协议边界，不把这些项目作为 baseline 发布阻塞。
-3. CLI 已观察到 managed 审计请求自动选择顶层 Skill、读取 Plugin Rule，并在修正 description 读取顺序后通过独立 Cursor 复审。该证据只证明当前本机路径，不承诺每个普通请求、所有模型或所有机器都会自动选择。
+3. CLI 已观察到方案审计请求自动选择顶层 Skill、读取 Plugin Rule，并在修正 description 读取顺序后通过独立 Cursor 复审。该证据只证明当前本机路径，不承诺每个普通请求、所有模型或所有机器都会自动选择。
 
 Cursor CLI 不会自行加载用户 Plugin Rule；Sopify 通过默认安装的顶层 Skill 显式读取该 Rule，不依赖 `--plugin-dir`、项目 Rule 或 prompt injector。Skill 发现和模型选择仍是宿主行为，因此 Doctor 继续把安装事实与 `cursor_cli_behavior` 黑盒证据分开。
 
-一次临时仓库黑盒已观察到四步读链、managed develop 与 `sopify_writer` 写回；本机顶层 Skill 复审又观察到 managed 请求自动选择和 Rule 消费。显式 finalize、跨模型一致性与可移植黑盒仍未完整认证；Doctor 的行为项继续保持静态 `BLACK_BOX_NOT_VERIFIED`。
+一次临时仓库黑盒已观察到四步读链、受管开发与 `sopify_writer` 写回；本机顶层 Skill 复审又观察到方案类请求按需选择入口并读取 Rule。显式 finalize、跨模型一致性与可移植黑盒仍未完整认证；Doctor 的行为项继续保持静态 `BLACK_BOX_NOT_VERIFIED`。
 
 官方依据：[`Cursor Plugins`](https://cursor.com/docs/plugins)、[`Plugin format`](https://cursor.com/docs/reference/plugins)、[`Cursor Rules`](https://cursor.com/docs/rules)、[`Cursor Agent Skills`](https://cursor.com/docs/skills)、[`Cursor Hooks`](https://cursor.com/docs/hooks)、[`Cursor CLI`](https://cursor.com/docs/cli/using)。
diff --git a/docs/getting-started.md b/docs/getting-started.md
index 74b6eb7..68010a4 100644
--- a/docs/getting-started.md
+++ b/docs/getting-started.md
@@ -141,9 +141,10 @@ release.
 
 ### Cursor IDE and local Agent CLI
 
-Cursor IDE uses one user-level Plugin rule. The same installation provides a
-managed-first Agent CLI entry, the shared Skill/payload tree under `~/.cursor/`,
-and user-level hooks in `~/.cursor/hooks.json`:
+One Cursor installation configures both local surfaces: the IDE gets a
+user-level Plugin Rule, while Agent CLI gets a top-level Skill for plan,
+development, resume, and closeout requests. Both share the phase Skills,
+payload under `~/.cursor/`, and user Hooks in `~/.cursor/hooks.json`:
 
 ```bash
 curl -fsSL https://github.com/evidentloop/sopify/releases/latest/download/install.sh | bash -s -- --target cursor:en-US
@@ -157,12 +158,13 @@ created during install. `--with-evidentloop` is not supported for Cursor in
 this release. The Plugin Rule uses valid `.mdc` frontmatter with
 `alwaysApply: true` and routes to `~/.cursor/skills/sopify`.
 
-Agent CLI does not load the user Plugin Rule by itself. The installed top-level
-`~/.cursor/skills/sopify/SKILL.md` is a managed-first, model-selected entry that
-reads the Rule before routing phase Skills. Selection is best-effort; ordinary
-questions do not have to trigger it, and Cloud Agent remains outside this
-release. No launcher, project Rule, prompt injector, or runtime is added. See
-[Cursor host acceptance](./cursor-host.md). The tier remains `BASELINE_SUPPORTED`.
+Agent CLI does not load the user Plugin Rule by itself. For a Sopify workflow
+request, Cursor may select `~/.cursor/skills/sopify/SKILL.md`; that entry reads
+the Rule before routing phase Skills. Ordinary questions do not have to use
+this path, and selection is not promised to behave identically across models.
+Cloud Agent remains outside this release. No launcher, project Rule, prompt
+injector, or runtime is added. See [Cursor host acceptance](./cursor-host.md).
+The tier remains `BASELINE_SUPPORTED`.
 
 ## Verify Setup
 
diff --git a/docs/how-sopify-works.en.md b/docs/how-sopify-works.en.md
index c7969b0..078f51e 100644
--- a/docs/how-sopify-works.en.md
+++ b/docs/how-sopify-works.en.md
@@ -109,7 +109,7 @@ Layer notes:
 | Claude | PROTOCOL_VERIFIED | `install.sh --target claude:en-US` | Full capability continuation |
 | Qoder | PROTOCOL_VERIFIED | `install.sh --target qoder` | Validated on Qoder CLI |
 | Copilot | BASELINE_SUPPORTED | `install.sh --target copilot` | Prompt-only; payload uplift planned |
-| Cursor | BASELINE_SUPPORTED | `install.sh --target cursor:en-US` | IDE user Plugin + managed-first CLI entry + shared Skills/payload/hooks |
+| Cursor | BASELINE_SUPPORTED | `install.sh --target cursor:en-US` | One install for IDE and local CLI; plan/development/resume/closeout entry is model-selected |
 
 ## Appendix: Plan Lifecycle
 
diff --git a/docs/how-sopify-works.md b/docs/how-sopify-works.md
index e187b06..ae97b63 100644
--- a/docs/how-sopify-works.md
+++ b/docs/how-sopify-works.md
@@ -109,7 +109,7 @@ Sopify 不默认安装或运行验证器。只用 Sopify、使用其他 Verifier
 | Claude | PROTOCOL_VERIFIED | `install.sh --target claude:zh-CN` | 全能力接续 |
 | Qoder | PROTOCOL_VERIFIED | `install.sh --target qoder` | 已在 Qoder CLI 验证 |
 | Copilot | BASELINE_SUPPORTED | `install.sh --target copilot` | 仅 prompt；payload 升级计划中 |
-| Cursor | BASELINE_SUPPORTED | `install.sh --target cursor:zh-CN` | IDE 用户 Plugin + managed-first CLI 入口 + 共享 Skills/payload/Hooks |
+| Cursor | BASELINE_SUPPORTED | `install.sh --target cursor:zh-CN` | 一次安装支持 IDE 与本地 CLI；CLI 方案/开发/接续/收口入口按需触发 |
 
 ## 附录：Plan 生命周期
 
diff --git a/index.html b/index.html
index 4ebb905..49c3b29 100644
--- a/index.html
+++ b/index.html
@@ -252,7 +252,7 @@ <h2>The short version</h2>
           </details>
           <details>
             <summary>Does switching hosts give every host identical capabilities?</summary>
-            <p>Codex, Claude, and Qoder have protocol-verified paths. Copilot and Cursor are baseline-supported. One Cursor install provides the IDE Plugin Rule and a best-effort managed entry for local Agent CLI; Cloud Agent remains out of scope.</p>
+            <p>Codex, Claude, and Qoder have protocol-verified paths. Copilot and Cursor are baseline-supported. One Cursor install covers the IDE and local CLI; the CLI selects the plan, development, resume, and closeout entry when the request calls for it. Cloud Agent remains out of scope.</p>
           </details>
           <details>
             <summary>Is EvidentLoop required?</summary>
diff --git a/skills/en/cursor-plugin-readme.md.template b/skills/en/cursor-plugin-readme.md.template
index cb045af..c484ba7 100644
--- a/skills/en/cursor-plugin-readme.md.template
+++ b/skills/en/cursor-plugin-readme.md.template
@@ -1,20 +1,20 @@
 # Sopify
 
-Sopify is an adaptive workflow layer for Cursor IDE and local Agent CLI. It routes each request to the smallest appropriate path—read-only consultation, quick fix, analysis, design, or managed development—while keeping evidence and project-state boundaries explicit.
+One installation lets Sopify serve both Cursor IDE and local Agent CLI. The IDE loads the user-level Rule automatically; the CLI selects a top-level Skill for plan, development, resume, and closeout requests, then reads the same Rule.
 
 ## Use Sopify
 
-Describe the goal naturally. Sopify uses the current repository's instructions and `.sopify/` context, then loads only the relevant Skill. For explicit phase control, invoke the installed `/analyze`, `/design`, or `/develop` Skill.
+- **In the IDE**: describe the goal naturally. The Always Rule classifies intent and loads the phase Skills that are needed.
+- **In the CLI**: describe a plan, development, or resume goal directly. Ordinary questions do not have to enter Sopify; for explicit phase control, `/analyze`, `/design`, and `/develop` remain available.
 
-## How it works
+## Shared capabilities
 
-- **IDE Always Rule** classifies intent and routes requests in Cursor IDE.
-- **CLI top-level Skill** reads the same Rule for managed requests in local Agent CLI; model selection is best-effort.
-- **Skills** provide the phase-specific procedures.
-- **User Hooks** guard protocol state and receipt files against obvious direct writes; they are not a security sandbox.
+- **Phase Skills** provide analysis, design, development, knowledge-base, and template procedures.
+- **Payload** provides the versioned protocol and `sopify_writer`.
+- **User Hooks** reject obvious direct writes to protocol state, handoff, and receipt files; they are not a security sandbox.
 
 ## Support boundary
 
-- Cursor IDE uses the Always Rule. Cursor Agent CLI installs a managed-first top-level Skill by default and reuses the same phase Skills, payload, and user Hooks.
-- Support remains `BASELINE_SUPPORTED`: IDE and CLI evidence stay separate, and Sopify does not promise that every ordinary request selects the top-level Skill, Cloud Agents, or identical behavior across models.
-- Sopify does not modify Cursor settings, proxy, model, account, API keys, keychain, or MCP configuration. Cloud Agents are out of scope.
+- Support remains `BASELINE_SUPPORTED`; IDE and CLI behavior evidence stays separate.
+- The model selects the CLI entry based on the request. Identical behavior is not promised for ordinary questions, every model, or every machine.
+- Sopify does not modify Cursor settings, proxy, model, account, API keys, keychain, or MCP. Cloud Agents are out of scope.
diff --git a/skills/zh/cursor-plugin-readme.md.template b/skills/zh/cursor-plugin-readme.md.template
index 2d8a024..d085eb4 100644
--- a/skills/zh/cursor-plugin-readme.md.template
+++ b/skills/zh/cursor-plugin-readme.md.template
@@ -1,20 +1,20 @@
 # Sopify
 
-Sopify 是 Cursor IDE 与本地 Agent CLI 的自适应工作流层。它先把请求路由到最小适用路径——只读咨询、快速修复、需求分析、方案设计或受管开发——并明确约束证据和项目状态边界。
+一次安装即可让 Sopify 同时服务于 Cursor IDE 与本地 Agent CLI。IDE 会自动加载用户级 Rule；CLI 会在方案、开发、继续和收口等请求中按需选择顶层 Skill，再读取同一 Rule。
 
 ## 使用方式
 
-直接用自然语言说明目标。Sopify 会结合当前仓库的项目指令与 `.sopify/` 上下文，只加载相关 Skill。需要明确指定阶段时，可以调用已安装的 `/analyze`、`/design` 或 `/develop` Skill。
+- **在 IDE 中**：直接用自然语言说明目标，Always Rule 会分类意图并加载需要的阶段 Skill。
+- **在 CLI 中**：直接描述方案、开发或接续目标。普通问答不强制进入 Sopify；需要明确指定阶段时，仍可调用 `/analyze`、`/design` 或 `/develop`。
 
-## 工作方式
+## 共用能力
 
-- **IDE Always Rule**：在 Cursor IDE 中分类意图并路由请求。
-- **CLI 顶层 Skill**：在本地 Agent CLI 中为 managed 请求读取同一 Rule；模型选择属于 best-effort。
-- **Skills**：提供各阶段的执行细则。
-- **用户级 Hooks**：对协议 state、handoff 与 receipt 提供明显直写保护；它们不是安全沙箱。
+- **阶段 Skills**：提供分析、设计、开发、知识库和模板细则。
+- **Payload**：提供版本化协议与 `sopify_writer`。
+- **用户级 Hooks**：拒绝对协议 state、handoff 与 receipt 的明显直接写入；它们不是安全沙箱。
 
 ## 支持边界
 
-- Cursor IDE 使用 Always Rule；Cursor Agent CLI 默认安装 managed-first 顶层 Skill，并复用同一阶段 Skills、payload 与用户 Hooks。
-- 当前支持档位为 `BASELINE_SUPPORTED`；IDE 与 CLI 证据分开记录，不承诺每个普通请求都选择顶层 Skill、Cloud Agent 或跨模型完全一致。
-- Sopify 不修改 Cursor settings、代理、模型、账号、API Key、钥匙串或 MCP 配置；不支持 Cloud Agent。
+- 当前支持档位为 `BASELINE_SUPPORTED`；IDE 与 CLI 的行为证据分开记录。
+- CLI 入口由模型按请求选择，不保证普通问答、所有模型或所有机器表现完全一致。
+- Sopify 不修改 Cursor settings、代理、模型、账号、API Key、钥匙串或 MCP；不支持 Cloud Agent。
diff --git a/tests/test_installer_status_doctor.py b/tests/test_installer_status_doctor.py
index 3f859e1..6745bae 100644
--- a/tests/test_installer_status_doctor.py
+++ b/tests/test_installer_status_doctor.py
@@ -134,8 +134,9 @@ def test_cursor_install_proves_split_paths_without_touching_settings(self) -> No
             self.assertTrue(readme.is_file())
             manifest_payload = json.loads(manifest.read_text(encoding="utf-8"))
             self.assertNotIn("rules", manifest_payload)
-            self.assertIn("adaptive workflow layer", readme.read_text(encoding="utf-8"))
-            self.assertIn("BASELINE_SUPPORTED", readme.read_text(encoding="utf-8"))
+            readme_text = readme.read_text(encoding="utf-8")
+            self.assertIn("Cursor IDE and local Agent CLI", readme_text)
+            self.assertIn("BASELINE_SUPPORTED", readme_text)
             self.assertTrue((skill_root / "SKILL.md").is_file())
             self.assertTrue((skill_root / "analyze" / "SKILL.md").is_file())
             self.assertTrue((skill_root / "design" / "SKILL.md").is_file())
diff --git a/zh-CN.html b/zh-CN.html
index 9cedca9..e9f5cbf 100644
--- a/zh-CN.html
+++ b/zh-CN.html
@@ -245,7 +245,7 @@ <h2>先说最重要的</h2>
           </details>
           <details>
             <summary>换宿主后能力完全一样吗？</summary>
-            <p>Codex、Claude 和 Qoder 已通过协议路径验证。Copilot 和 Cursor 为基线支持；一次 Cursor 安装同时提供 IDE Plugin Rule 与本地 Agent CLI 的 best-effort managed 入口，Cloud Agent 仍不在范围内。</p>
+            <p>Codex、Claude 和 Qoder 已通过协议路径验证。Copilot 和 Cursor 为基线支持；Cursor 安装一次即可覆盖 IDE 与本地 CLI，CLI 会在方案、开发、接续和收口类请求中按需选择入口。Cloud Agent 仍不在范围内。</p>
           </details>
           <details>
             <summary>必须安装 EvidentLoop 吗？</summary>

From e1a91a360d0c0a76b54d147fdf5e0316812e75df Mon Sep 17 00:00:00 2001
From: "sanze.li" <weixin.li@nio.com>
Date: Sat, 29 Aug 2026 23:51:43 +0800
Subject: [PATCH 3/3] test: align Cursor README contract

---
 tests/test_distribution.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/test_distribution.py b/tests/test_distribution.py
index 4403922..9e2a9b6 100644
--- a/tests/test_distribution.py
+++ b/tests/test_distribution.py
@@ -249,7 +249,7 @@ def test_cursor_distribution_reports_user_plugin_without_project_write(self) ->
             readme_path = rule_path.parent.parent / "README.md"
             self.assertIn(f"用户 Plugin 规则：{rule_path}", rendered)
             self.assertTrue(readme_path.is_file())
-            self.assertIn("自适应工作流层", readme_path.read_text(encoding="utf-8"))
+            self.assertIn("Cursor IDE 与本地 Agent CLI", readme_path.read_text(encoding="utf-8"))
             self.assertIn("未修改任何项目目录", rendered)
             self.assertIn("Developer: Reload Window", rendered)
             self.assertIn("确认 Sopify Plugin 及其 Always Rule 已启用", rendered)