# v1.3.5 Release Notes

> v1.3.4 -> v1.3.5

## Summary

Separates per-request model choices from session defaults, so the web chat keeps its picks in the browser and a removed model no longer strands a session. Widens the always-loaded tool set and slims the system prompt so the agent reuses earlier results instead of redoing work. Chatbot channels now steer a running task and start a new one only on request.

<details>
<summary>翻譯</summary>

把單次請求的模型選擇與 session 預設分開：Web 對話的選擇留在瀏覽器，移除模型也不會讓 session 卡住。擴充常駐工具並精簡 system prompt，讓 agent 沿用先前結果而不是重做。聊天機器人頻道改為 steer 執行中的任務，只在要求時另開新任務。

</details>

## Changes

### FEAT

- Add `/new` to Telegram and Discord: while a task is running, other messages steer it silently, and `/new <message>` (also `/new@<bot>`) starts a separate task; Telegram registers `/new` in the bot command menu on startup (@pardnchiu) [4784593b]
- Add per-request `reasoning` to `/v1/send`; `model` and `reasoning` apply to that request only, an omitted `model` uses the session model, and `auto` always goes to the dispatcher (@pardnchiu) [057a0922]
- Keep the web chat's model and reasoning picks in browser cookies, sent with each request and preferred over the session default when still available, instead of writing them to the session (@pardnchiu) [057a0922]
- Add a `figure` topic to `reasoning_guide` holding chart and diagram examples plus the extended Markdown syntax; the system prompt points to it whenever an answer needs a chart, flowchart, diagram or formula (@pardnchiu) [542b6e50]
- Load `chat_history`, `ask_user` and `write_result` with full schemas alongside `find_tools`, `run_tool`, `reasoning_guide` and `write_todo`, so follow-up tasks read earlier tool results and reports instead of fetching again (@pardnchiu) [542b6e50]
- Append a `next:` hint to failed tool results, pointing to `store_secret` for missing or rejected credentials and to `error_history` once a different fix works (@pardnchiu) [542b6e50]
- Auto-scroll every web list to its selected item (@pardnchiu) [fe6c6e8c]

<details>
<summary>翻譯</summary>

- Telegram 與 Discord 新增 `/new`：任務執行中其他訊息一律靜默 steer 該任務，`/new <訊息>`（也接受 `/new@<bot>`）另開任務；Telegram 啟動時把 `/new` 註冊進指令選單
- `/v1/send` 新增單次的 `reasoning`；`model` 與 `reasoning` 只作用於該次請求，未帶 `model` 用 session 模型，`auto` 一律交給 dispatcher
- Web 對話的模型與 reasoning 選擇改存瀏覽器 cookie，隨每次請求送出，值仍可選時優先於 session 預設，不再寫入 session
- `reasoning_guide` 新增 `figure` topic，收錄圖表範例與 Markdown 擴充語法；回答需要圖表、流程圖、圖解或公式時，system prompt 指向該 guide
- `chat_history`、`ask_user`、`write_result` 與 `find_tools`、`run_tool`、`reasoning_guide`、`write_todo` 一同以完整 schema 載入，接續任務會讀取先前的工具結果與報告，不再重抓
- 工具失敗結果附加 `next:` 提示：憑證缺漏或被拒時指向 `store_secret`，改用其他方法成功後指向 `error_history`
- Web 所有清單自動捲動到選中項目

</details>

### UPDATE

- Reset sessions that use a model to `auto` when that model is removed, from both the TUI and `DELETE /v1/models/*name` (@pardnchiu) [057a0922]
- Reset the session default to `auto` on pending resume only when the resumed model equals the default and no longer exists; other cases leave the default untouched and route normally (@pardnchiu) [057a0922]
- Rewrite the descriptions and parameter docs of the seven default tools, and drop the `mcp` and `system` parameters from `find_tools` (@pardnchiu) [542b6e50, 057a0922]
- Shorten the system prompt by moving Markdown syntax into the `figure` guide and condensing the delivery rules (@pardnchiu) [542b6e50]
- Move the TUI's chatbot token-verification notice to the debug log (@pardnchiu) [4784593b]
- Add bottom padding to sidebar lists (@pardnchiu) [fe6c6e8c]

<details>
<summary>翻譯</summary>

- 移除模型時（TUI 與 `DELETE /v1/models/*name`），使用該模型的 session 改回 `auto`
- pending 恢復時，只有恢復的模型等於 default 且已不存在才把 session default 改回 `auto`；其他情況不動 default，交給一般路由
- 改寫 7 個預設工具的描述與參數說明，`find_tools` 移除 `mcp` 與 `system` 參數
- 精簡 system prompt：Markdown 語法移入 `figure` guide，交付規則縮短
- TUI 的聊天機器人 token 驗證提示改寫入 debug log
- 側欄清單加上底部留白

</details>

### CHORE

- Bump `github.com/pardnchiu/go-bot` to v0.5.1 for command menu registration (@pardnchiu) [4784593b]

<details>
<summary>翻譯</summary>

- 升級 `github.com/pardnchiu/go-bot` 至 v0.5.1，以支援指令選單註冊

</details>

## Scope

- `internal/runtime/chatbot/` — FEAT (`chatbot.go`, `discord/run.go`, `telegram/new.go`, `telegram/run.go`)
- `internal/agents/exec/` — FEAT, UPDATE (`run.go`, `toolCall.go`, `model/route.go`, `model/select.go`)
- `internal/runtime/routes/handler/` — FEAT, UPDATE (`send.go`, `model.go`)
- `internal/session/config/bot/` — UPDATE (`bot.go`)
- `internal/tools/` — FEAT, UPDATE (multiple files)
- `internal/runtime/tui/` — UPDATE (`update.go`, `commnadModelRemove.go`)
- `configs/` — FEAT, UPDATE (`configs.go`, `prompts/guide/figure.md`, `prompts/system_prompt/system_prompt.md`)
- `page/src/`, `page/public/` — FEAT, UPDATE (multiple files)
- `go.mod`, `go.sum` — CHORE

---

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