DOC: Clarify clean-context plan execution

The Plan Mode documentation described skills as simply omitted and did not explain the authority, provenance, resumability, or file-history behavior of a clean implementation handoff.

Document the user-role plan submission, current compact skill rediscovery, planned and implementation badges, unchanged workspace, resumable source session, and best-effort checkpoint inheritance. Duplicating full skill documents in the handoff was rejected because that would defeat the clean-context guarantee.

Verified both language variants against the implemented flow; git diff --check, npm run check, and npm test pass.

Co-authored-by: Codex <noreply@openai.com>

Author: cmp0xff

docs/plan-mode.md

@@ -92,9 +92,15 @@ Plan Mode 的核心规则是**只规划，不动手**。例如以下操作是**
 | **1. implement this plan** | 退出 Plan Mode，自动发送实现指令，让 AI 开始按方案写代码 |
 | **2. stay in Plan mode** | 保持在 Plan Mode，继续修改或完善方案 |
 | **3. switch to Default mode** | 退出 Plan Mode，回到默认模式（不自动开始实现） |
-| **4. clear context and implement this plan** | 派生一个干净的新会话（仅携带系统提示、运行上下文、AGENTS.md 指令与方案），在全新上下文中开始实现 |
+| **4. clear context and implement this plan** | 使用当前系统提示、运行上下文、AGENTS.md 指令及仍可用 skills 的精简目录派生干净会话，再把已批准方案作为第一条用户消息提交 |
 
-注意：该选项不会把规划阶段加载的 skills 带入新会话；实现时可按需通过 skill 工具重新加载。
+该交接不会复制规划对话、工具输出或完整 skill 文档。规划阶段曾展示或加载的 skills 会按当前配置重新扫描，
+只保留仍存在的名称和最新描述。完整交接指令与方案以用户角色提交，因此会再次经过正常的 skill 自动匹配，
+实现会话也能按通常方式加载匹配的 skill。
+
+工作区内容不会改变。已完成的源会话会保留并标记为 `planned`，可随时恢复；干净会话标记为
+`implementation`。文件历史按最佳努力继承：若源会话已有 checkpoint，实现会话会从该 checkpoint 开始，
+以便继续使用 `/undo`；文件历史仓库缺失则不会阻止交接。
 
 你可以用数字键 `1-4` 直接选择，也可以用 `↑/↓` 移动光标后按 `Enter` 确认。按 `Esc` 等同于选择 "stay in Plan mode"。
 

docs/plan-mode_en.md

@@ -92,9 +92,17 @@ After the plan is output, Deep Code automatically shows a choice dialog—no ext
 | **1. implement this plan** | Leave Plan Mode and automatically send an implementation prompt so the AI starts coding |
 | **2. stay in Plan mode** | Stay in Plan Mode to continue refining the plan |
 | **3. switch to Default mode** | Leave Plan Mode and return to Default mode without starting implementation |
-| **4. clear context and implement this plan** | Derive a clean new session (carrying only the system prompt, runtime context, AGENTS.md instructions, and the plan) to implement in a fresh context |
+| **4. clear context and implement this plan** | Derive a clean session with the current system prompt, runtime context, AGENTS.md instructions, and a compact catalog of still-available skills; then submit the approved plan as its first user message |
 
-Note: skills loaded during planning are not carried over; implementation can reload them via the skill tool as needed.
+The handoff does not copy planning dialogue, tool output, or full skill documents. Skills advertised or loaded while
+planning are rediscovered from the current configuration, and only their current names and descriptions are retained.
+Because the complete handoff and plan are submitted in the user role, normal automatic skill matching runs again and
+the implementation session can load any matching skill normally.
+
+The workspace is unchanged. Deep Code keeps the completed source session available to resume and marks it `planned`;
+the clean session is marked `implementation`. File-history inheritance is best effort: when the source has a checkpoint,
+the implementation session starts from it so `/undo` remains useful, while a missing file-history repository does not
+block the handoff.
 
 You can press `1-4` to select directly, or use `↑/↓` to move the cursor and `Enter` to confirm. Pressing `Esc` is equivalent to choosing "stay in Plan mode."