diff --git a/AGENTS.md b/AGENTS.md
index 90e5553..32be2f5 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -43,6 +43,15 @@ Use it whenever work requires any of:
 
 Do not treat non-chat work as ordinary reply flow.
 
+### Reply Closure Rule
+
+On Telegram, if the final actionable part of your reply needs the human to decide, confirm, approve, stop, continue, rerun, or choose a next step:
+- do **not** end with plain text only
+- do **not** say buttons will be used unless you are actually sending them
+- either send real inline buttons or execute the most reasonable next step directly
+
+If you fail this, call it a workflow violation and correct it immediately.
+
 ## Memory
 
 You wake up fresh each session. These files are your continuity:
diff --git a/WORKFLOW.md b/WORKFLOW.md
index c70ac16..8618d4a 100644
--- a/WORKFLOW.md
+++ b/WORKFLOW.md
@@ -119,6 +119,18 @@ Subagent 指派後 **5 分鐘內若無結果**：
 - 若需要總管同意修改，應直接用按鈕送出 `OK / 先看改法` 之類的選項
 - 若總管已明確表示「就是要你修」，優先直接進入修復流程，不要再把修復方案包成純文字選單
 
+### Reply Closure Button Gate
+- 在 Telegram 上，只要**回覆的最後可執行部分**需要總管做選擇、確認、批准、停止、繼續、重跑、收下或決定下一步，就不能只用普通文字結尾。
+- 這時只能做兩件事：
+  1. **真的送出 inline buttons**
+  2. 若其實不需要總管決策，**直接執行最合理的下一步**
+- 「文字裡說會用按鈕」但沒有實際送出按鈕，視為**同樣違規**。
+- 這條 gate 特別適用於：
+  - long-task checkpoint 收尾
+  - 測試結果判定
+  - accept / rerun / stop 類互動
+  - approval / confirm 類收尾
+
 ### Two-phase gate（硬閘門：先報備→再執行）
 以下動作一律視為「對外 / 非瑣碎」：
 - 發 Lobby 訊息（message tool send）
diff --git a/WORKFLOW_GATES.md b/WORKFLOW_GATES.md
new file mode 100644
index 0000000..23498b5
--- /dev/null
+++ b/WORKFLOW_GATES.md
@@ -0,0 +1,41 @@
+# WORKFLOW_GATES.md
+
+## Reply-Closure Button Gate
+
+When operating on Telegram, if the **final actionable part** of a reply requires the user to choose, confirm, approve, stop, continue, or select a next step, the assistant must not end with plain text alone.
+
+### Hard rule
+
+If the reply ends in any owner decision gate, the assistant must do one of two things:
+1. **Send real inline buttons**
+2. **Execute the most reasonable next step directly** if no real decision is needed
+
+### Forbidden behavior
+
+These are violations when used as the closing interaction on Telegram:
+- `1 / 2 / 3`
+- `A / B / C`
+- `請回我 1`
+- `如果你要，我可以...`
+- `要不要我繼續？`
+- saying buttons will be used, but not actually sending them
+
+### Required interpretation
+
+The gate applies to:
+- long-task checkpoints
+- test progress summaries
+- approval requests
+- next-step choices
+- accept / rerun / stop style decisions
+
+### Violation standard
+
+If the assistant reaches a user-decision closure and no real inline buttons were delivered, treat it as a workflow violation even if the reply mentioned buttons in text.
+
+### Corrective rule
+
+If this violation happens:
+- acknowledge the violation plainly
+- immediately send the real button message
+- record the lesson into workflow / memory if it exposed a missing rule
diff --git a/memory/2026-04-22.md b/memory/2026-04-22.md
index 99b92d3..962bbb4 100644
--- a/memory/2026-04-22.md
+++ b/memory/2026-04-22.md
@@ -48,4 +48,5 @@
 - skill 目標：把 `general_chat vs long_task` 判準、狀態機、checkpoint 規格、no-fake-progress、stop-clock gate 收斂到外掛式 workflow layer。
 - 已同步更新 `WORKFLOW.md`，要求：只要不是 ordinary single-turn general chat，就必須套用 `long-task-governor`。
 - live test 期間實際抓到一個缺口：在 Telegram 上做 long-task checkpoint / next-step 收尾時，我仍可能退回純文字 `1 / 2 / 3` 選單。
-- 因此已再補一條 skill 規則：`Telegram interaction guard`，禁止 governed long-task flow 在 Telegram 上用純文字選單收尾；若需要選擇，必須改用 inline buttons，否則直接執行最合理下一步。
+- 因此已先補一條 skill 規則：`Telegram interaction guard`，禁止 governed long-task flow 在 Telegram 上用純文字選單收尾；若需要選擇，必須改用 inline buttons，否則直接執行最合理下一步。
+- 後續再往前補強到更直接的操作層：新增 `Reply Closure Button Gate` 概念，明確規定只要回覆最後的可執行部分需要總管決定、確認、批准、停止、繼續、重跑或選下一步，就不能只在文字裡說會用按鈕，必須真的送出 inline buttons，或直接執行最合理下一步。