# v1.1.2 Release Notes

> v1.1.1 -> v1.1.2

## Summary

Extends claude-code model discovery with context-window labels in the model picker. Consolidates scattered constants, timeouts and limits into one place with new defaults, and moves follow-up generation out of the event stream. Drops legacy data migrations.

<details>
<summary>翻譯</summary>

擴充 claude-code 模型探索，並於模型選擇器顯示 context window。將分散的常數、逾時與上限集中管理並調整預設值，後續建議生成移出事件串流。移除舊版資料遷移。

</details>

## Changes

### FEAT

- Add claude-code model discovery and show each model's context window in the model picker; the claude-code entry in model add is hidden unless claude-code is enabled (@pardnchiu) [c64765e4, 60119a26]

<details>
<summary>翻譯</summary>

- 新增 claude-code 模型列表查詢，並於模型選擇器顯示各模型的 context window；未啟用 claude-code 時，新增模型選單不顯示該項目

</details>

### UPDATE

- Persist claude-code sessions only under the 1h cache TTL; 5m and one-shot processes no longer write state or resume, to keep cache hits (@pardnchiu) [bdd9adb2]
- Serve follow-up title and suggestions from a separate `POST /v1/session/:session_id/followup` endpoint instead of the `/v1/send` event stream; only the WebUI calls it, so other entry points no longer receive them (@pardnchiu) [a1c0b4ab]
- Centralize constants, timeouts, retry limits and size caps in `configs`, with changed defaults: default tool timeout 15 minutes, run_command timeout 30 minutes, watch-script depth 4, one extra retry attempt, follow-up limits reduced, history messages 16, tool iterations 256, session tasks fixed at 3. Config limits can now only raise the defaults and defaults are no longer written back (@pardnchiu) [8c43578a]
- Register per-API tool timeouts with the tool registry so each API tool's configured timeout takes effect (@pardnchiu) [8c43578a]
- Raise the pending snapshot truncation caps to 32 KiB for results and 4 KiB for args, and tell the model to re-call the tool when elided output matters (@pardnchiu) [8c43578a]

<details>
<summary>翻譯</summary>

- claude-code session 僅在 1 小時快取 TTL 下持久化；5 分鐘與一次性程序不再寫入狀態或 resume，以維持快取命中
- 後續建議與標題改由獨立的 `POST /v1/session/:session_id/followup` 端點提供，不再走 `/v1/send` 事件串流；僅 WebUI 會呼叫，其他入口不再收到
- 將常數、逾時、重試與大小上限集中至 `configs` 並調整預設值：預設工具逾時 15 分鐘、run_command 逾時 30 分鐘、監看腳本深度 4、重試多一次、followup 上限縮小、歷史訊息 16 則、工具迭代 256 次、session 並行任務固定為 3。設定檔的上限只能調高預設值，預設值也不再回寫設定檔
- 將各 API 工具的逾時註冊至工具登錄表，使各工具設定的逾時確實生效
- 將 pending 快照截斷上限提高至 result 32 KiB、args 4 KiB，並提示模型在被省略的內容重要時重新呼叫工具

</details>

### REMOVE

- Remove legacy data migrations and the path helpers and parsers they relied on; data from older versions (`bot.json`, `config.json`, `usage.log`, `history/`, `summary.meta.json`, old DB schema) is no longer upgraded automatically, so upgrade through a version that still carries the migrations first. Obsolete config-key warnings and `planner_model` cleanup go with them (@pardnchiu) [09a3e37e]

<details>
<summary>翻譯</summary>

- 移除舊版資料遷移與其依賴的路徑函式、解析邏輯；舊版資料（`bot.json`、`config.json`、`usage.log`、`history/`、`summary.meta.json`、舊 DB schema）不再自動升級，須先經過仍含遷移的版本。過時設定 key 的警告與 `planner_model` 清除一併移除

</details>

### CHORE

- Bump go-llm-router to v0.8.2 (@pardnchiu) [c64765e4]

<details>
<summary>翻譯</summary>

- 升級 go-llm-router 至 v0.8.2

</details>

## Scope

- `internal/agents/claudeCode/` — FEAT, UPDATE (`models.go`, `agent.go`, `process.go`)
- `internal/runtime/tui/` — FEAT, UPDATE (`commandModelAdd.go`, `commandUsage.go`)
- `configs/` — UPDATE (`configs.go`)
- `internal/filesystem/` — UPDATE, REMOVE (`runtime.go`, `path.go`)
- `internal/runtime/routes/` — UPDATE (`new.go`, `handler/sendFollowup.go`)
- `internal/runtime/toolAdapter/api/` — UPDATE (`adapter.go`, `execute.go`)
- `internal/tools/` — FIX, UPDATE (`runCommandWatch.go`, `runCommand.go`, `executor.go`, `interactive/askUser.go`)
- `internal/runtime/store/` — REMOVE (`migrateAction.go`, `migrateSession.go`, `store.go`)
- `internal/session/` — REMOVE (`summary/migrate.go`, `usage/migrate.go`, `history/record.go`)
- `internal/agents/exec/` — UPDATE (multiple files)
- `page/src/js/` — UPDATE (`_event.js`)
- `README.md`, `doc/README.zh.md` — DOC
- `go.mod`, `go.sum` — CHORE

---

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