# v1.1.0 Release Notes

> v1.0.27 -> v1.1.0

## Summary

Adds a subscription-based model provider and moves reasoning selection from a global toggle to a per-session choice. Renames rules to roles while keeping the legacy names working, and reworks terminal popups into tabs.

<details>
<summary>翻譯</summary>

新增訂閱制模型供應商，並將 reasoning 選擇由全域開關改為逐 session 決定。規則改名為角色但保留舊名相容，終端介面的彈窗也改為分頁。

</details>

## Changes

### FEAT

- Add the claude-code provider, which runs the local Claude CLI as a model: opt-in at startup, sends tool names with only the `find_tools` schema, reads usage through the CLI without calling the API, lists models from `models.agenvoy.com` via go-llm-router v0.8.1, and ranks subscription providers first (@pardnchiu) [abf60707, 3fa1968b, e8886081, 59d40b9a, 852cc1b0]

<details>
<summary>翻譯</summary>

- 新增 claude-code provider，以本機 Claude CLI 作為模型：啟動時選擇啟用、只傳工具名稱與 `find_tools` schema、用量透過 CLI 取得而不打 API、模型清單經 go-llm-router v0.8.1 取自 `models.agenvoy.com`，並在排序中優先訂閱制 provider

</details>

### FIX

- Count cache writes in the input of `claude` usage, and stop the web usage totals from counting them twice (@pardnchiu) [434ae210, 91cb3fc6]
- Resolve `model: "auto"` on a subagent leg through the dispatcher instead of rejecting it, and let a session pinned to its own model override it; the `pass` tier description now states that fallback skips it too (@pardnchiu) [84c11270]

<details>
<summary>翻譯</summary>

- `claude` 用量的 input 計入 cache 寫入量，並修正 Web 用量總計重複計入
- subagent leg 的 `model: "auto"` 改由 dispatcher 選模而非拒絕，session 固定模型時以其覆寫；`pass` tier 說明改為註明 fallback 也會略過

</details>

### UPDATE

- Replace the global `auto_reasoning` toggle with a per-session choice driven by the selector's work kind, with `auto` as a session level; the config key and the `auto_reasoning` fields of `/v1/model` and `/v1/session/:id` are removed (@pardnchiu) [da03ee7e, 18f3a2f9, 5c8834aa]
- Fetch remaining provider quota asynchronously after completion instead of blocking the completion event, and add `/v1/model/quota` (@pardnchiu) [ddfff76c, 91cb3fc6]
- Require a model on every subagent leg and size legs one tier lower (@pardnchiu) [cdc9c9c8]
- Skip `ask_user` when the skill has no such step (@pardnchiu) [e2af78af]
- Cap the TypeSafe selector wait so a slow answer falls back to the dispatcher model (@pardnchiu) [9957b975]
- Show cache hit rates in usage statistics (@pardnchiu) [1d80ad73]
- Rework the TUI: tabbed channel and model-routing popups, channel admin moved into config, a persistent notice box, streamlined session actions and an animated plan indicator (@pardnchiu) [8052ca0b, 5c8834aa, 3e95fac7, 9957b975, 7da5ebb5, 3fa1968b]

<details>
<summary>翻譯</summary>

- 以逐 session 依 selector 工作類型決定 reasoning 等級（`auto` 為 session 等級之一）取代全域 `auto_reasoning` 開關；config 鍵，以及 `/v1/model` 與 `/v1/session/:id` 的 `auto_reasoning` 欄位一併移除
- 完成後非同步查詢 provider 剩餘額度，不再阻塞完成事件，並新增 `/v1/model/quota`
- subagent 每條 leg 必須指定模型，並整體低一階分級
- skill 沒有 `ask_user` 步驟時不詢問
- 限制 TypeSafe selector 等待時間，過慢時退回 dispatcher 模型
- 用量統計顯示 cache 命中率
- 重整 TUI：channel 與模型路由彈窗改為分頁、channel admin 移入 config、常駐 notice 框、精簡 session 操作、動態計畫指示

</details>

### REFACTOR

- Rename rules to roles across the UI, API and session storage while the legacy `rule` routes, fields and column keep working (@pardnchiu) [730f2a8b, 87d0197c, 1d80ad73]

<details>
<summary>翻譯</summary>

- UI、API 與 session 儲存中的 rule 全面改名為 role，舊的 `rule` 路由、欄位與資料欄仍可使用

</details>

## Scope

- `cmd/app/`, `internal/agents/claudeCode/` — FEAT
- `internal/agents/exec/` — FEAT, UPDATE, FIX
- `internal/runtime/routes/` — UPDATE, REFACTOR
- `internal/runtime/tui/` — UPDATE, REFACTOR
- `internal/runtime/store/` — REFACTOR
- `internal/session/` — UPDATE, FIX
- `internal/tools/subagent/` — UPDATE, FIX
- `configs/prompts/` — FEAT, UPDATE
- `page/` — UPDATE, REFACTOR, FIX
- `doc/` — DOC
- `go.mod` — CHORE

---

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