# v1.3.14 Release Notes

> v1.3.13 -> v1.3.14

## Summary

Telegram and Discord chats gain session controls, so a conversation can be steered, stopped, resumed and trimmed without returning to the terminal. Permission prompts no longer repeat across a resumed task, and reply footers now look the same on every channel.

<details>
<summary>翻譯</summary>

Telegram 與 Discord 聊天新增 session 控制指令，不必回到終端機就能調整、停止、接續與精簡對話。接續中的任務不再重複詢問權限，各頻道的回覆 footer 也統一為同一格式。

</details>

## Changes

### FEAT

- Add `/model` and `/reasoning` to Telegram and Discord: without an argument they open a menu of registered models or reasoning levels, with one they set it directly for the chat, and unknown models or invalid levels are rejected; `/new` now requires input and replies with its usage otherwise (@pardnchiu) [2871d35f]
- Add `/pending` to Telegram and Discord: a menu of interrupted tasks (`❓` marks those waiting for answers) where picking one resumes it, or re-sends its pending questions first (@pardnchiu) [6d214322]
- Add `/stop` to Telegram and Discord: a menu of the chat's running tasks where picking one cancels it along with its background work (@pardnchiu) [275d60bf]
- Add `/compact` to Telegram and Discord: asks for confirmation, shows a temporary `Compacting...` message while history is trimmed by LLM analysis, then reports how many messages were removed; refused while a task is running (@pardnchiu) [fdf3b07e]

<details>
<summary>翻譯</summary>

- Telegram 與 Discord 新增 `/model`、`/reasoning`：不帶參數時跳出已註冊模型或 reasoning 等級的選單，帶參數時直接為該聊天設定，模型不存在或等級不合法時拒絕；`/new` 必須附帶輸入，否則回覆用法
- Telegram 與 Discord 新增 `/pending`：以選單列出中斷的任務（`❓` 標示等待回答者），點選即接續，等待回答的任務會先重新送出問題
- Telegram 與 Discord 新增 `/stop`：以選單列出該聊天執行中的任務，點選即取消，連同其背景工作一起停止
- Telegram 與 Discord 新增 `/compact`：先詢問確認，以 LLM 分析精簡歷史期間顯示暫時的 `Compacting...` 訊息，完成後回報移除的訊息數；有任務執行中時拒絕

</details>

### FIX

- Fix repeated confirmations caused by the `$HOME` / sensitive-path rework in v1.3.7 [d5d3df29]: since then any sensitive path under `$HOME` (`.claude`, `.ssh`, `.git`...) forced a prompt even with always-allow on or after "allow this turn"; combined with "allow this turn" never carrying over to a resumed pending task, long tasks and resumed tasks asked again on every read. Now "allow this turn" is stored in the pending task and reused on resume, and with always-allow on `read_files` / `find_files` read those paths without prompting; writes, commands and paths outside `$HOME` keep their existing confirmation and password rules (@pardnchiu) [8ab1ee21]

<details>
<summary>翻譯</summary>

- 修正 v1.3.7 `$HOME`／敏感路徑改版 [d5d3df29] 造成的重複確認：自那版起，`$HOME` 內的敏感路徑（`.claude`、`.ssh`、`.git`...）即使開啟 always-allow 或選了「allow this turn」仍一律要求確認；加上「allow this turn」原本就不會帶到接續的 pending 任務，長任務與接續中的任務每次讀取都會再問一次。現在「allow this turn」寫進 pending 任務、接續時沿用；always-allow 開啟時 `read_files`／`find_files` 讀取這些路徑不再詢問；寫入、執行指令與 `$HOME` 外的路徑維持原本的確認與密碼規則

</details>

## Scope

- `internal/runtime/chatbot/` — FEAT, UPDATE (`command.go`, `chatbot.go`, `reply.go`, `telegram/`, `discord/`)
- `internal/tools/interactive/askUser.go` — FEAT, FIX
- `internal/agents/exec/` — FIX (`execute.go`, `toolCall.go`)
- `internal/utils/event.go`, `internal/runtime/tui/viewRender.go` — UPDATE
- `page/src/sass/right-content/chat/_index.scss`, `page/public/index.css` — UPDATE

---

Generated by [SKILL](https://github.com/pardnchiu/skill-version-generate)
