# v0.23.2 Release Notes

> v0.23.1 -> v0.23.2

## Summary

Brings inbound Telegram audio and video into the agent pipeline through a Gemini-backed transcription utility plus an auto-save attachment intake on the runtime side. Provider-dependent tools now gate on credential presence rather than model registration, so any agent can reach the utility surface without adding the backend as one of its own models.
<details>
<summary>翻譯</summary>
把 Telegram 音檔與影片接進 agent 流程：runtime 端自動存檔，Gemini 背書的轉文字 tool 接手後續解析。Provider-dependent tool 改以憑證存在為 gate，不再要求把後端加為使用者 agent，任何 agent 都能呼叫 utility surface。
</details>

## Changes

### FEAT
- Add Gemini media transcription tool and Telegram attachment intake (@pardnchiu) [1d9d1d7]

<details>
<summary>翻譯</summary>

- 新增 Gemini 背書的轉文字 tool 與 Telegram 附件自動接收

</details>

## Scope

- `internal/runtime/telegram/` — FEAT (`attachments.go`, `run.go`)
- `internal/agents/provider/gemini/stt/` — FEAT (`handler.go`, `register.go`)
- `internal/agents/provider/gemini/youtube/` — REFACTOR (`register.go`)
- `internal/agents/provider/openaiCodex/image2/` — REFACTOR (`register.go`)
- `internal/filesystem/` — REFACTOR (`reader.go`, removed `fileReader/csv.go`, `fileReader/image.go`)
- `internal/tools/file/` — UPDATE (`readFile.go`)
- `cmd/app/` — FEAT (`main.go`, `cmdDeamon.go`, `newTUI.go`)
- `go.mod`, `go.sum` — CHORE (`go-pkg v0.12.3`, `go-bot v0.2.1`)

***

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