SkillFoxx

bilibili-reply skill

A skill that analyses the video behind a Bilibili @ mention and posts a reply with clickable timestamps from your account.

  • 113 ★ on the repository
  • Last commit
  • MIT
Skill

Broad access

Why this level

  • The agent publishes comments in the name of your Bilibili account, in automatic mode without a check of each reply.
  • It needs a signed-in browser session that the agent fully controls.
All reasons and checks

bronyacccxz/dsh-bilibili-replyAuthor:bronyacccxz

Install

Works in

In your terminal, with SkillFoxx CLI

npx skillfoxx add skills/bilibili-reply

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • The agent publishes comments in the name of your Bilibili account, in automatic mode without a check of each reply.
  • It needs a signed-in browser session that the agent fully controls.

This is third-party code. Review the repository files before installing.

What it does

The bilibili-reply skill handles "@ me" notifications on Bilibili. The agent opens the notifications page in an already signed-in browser, checks the mention author's nickname against the list in `config.md`, finds the video and analyses it through the `video-watch` skill: local speech transcription and frames from the clip. The reply depends on the video type. For educational videos the agent writes a chaptered breakdown with `00:00` timestamps that Bilibili turns into links. For entertainment videos it leaves a short reaction without timestamps. The reply goes out through the reply button inside the notification so the author gets notified.

Who it is for. Bilibili creators and viewers on DeepSeek Harness who want to answer comment mentions without watching each video by hand.

Good fit when

  • A mention under a video asks a question and the reply needs links to moments in the clip.
  • An educational video calls for a chaptered breakdown with timestamps.
  • A song or clip without speech: the agent reruns transcription with the `--no-vad` flag to get text.
  • Mentions arrive constantly and checking them should go to a schedule: the `watch-at.ps1` script polls notifications every 30 seconds.

Not a fit when

  • The agent cannot control the browser where you are signed in to Bilibili.
  • Comments under videos must stay hand-written only.
  • The mention comes from someone other than the person running the skill: the check goes by nickname and other people's notifications are skipped.

Example request

I was mentioned on Bilibili. Check the notifications, analyse the video and reply with chapter timestamps.

Limitations

The skill expects DeepSeek Harness, browser automation such as BrowserSkill and the `video-watch` skill, which the repository does not include. Without a controllable browser signed in to Bilibili it does not work. It posts public comments from your account, so the author recommends a separate rarely used account. Frequent replies trigger a Bilibili check or error `-352`. Documentation and interface are in Chinese, and the ready scheduled run is described for Windows and PowerShell.

How to disable. Delete the `skills/bilibili-reply` folder from the DeepSeek Harness directory. If you set up automatic mode, remove the Windows scheduler task that `watch-at.ps1 -InstallTask` created.

FAQ

What has to be installed before the bilibili-reply skill?

DeepSeek Harness, `pnpm`, browser automation signed in to Bilibili (the author uses BrowserSkill) and the `video-watch` skill. The check is simple: the agent must open `https://message.bilibili.com/#/at` and read the notification list.

Why does the reply go through the notification instead of the field under the video?

A comment from the field under the video becomes a separate one and does not notify the author of the mention. The inline reply in the notification does.

Does the skill work without DeepSeek Harness?

The logic carries over: examples in `SKILL.md` use `browser_*` tools and you can swap in your own. The skill itself targets DSH.

What does the `config.md` file do?

It stores your nickname and the bot's nickname. The skill compares the mention author with these names and does not reply if they differ, so the setup step cannot be skipped.

How do I turn on automatic mode?

In the `tools` folder run `.\watch-at.ps1 -Init`, then `.\watch-at.ps1 -InstallTask`. The scheduler task checks mentions every 30 seconds and calls the model only when it replies.

Discussion

No questions or reviews yet.

Skills

A Bilibili CLI gives agents video search, subtitles and tools to manage posts and interactions on your own account

CLIBroad access383

Bili Note builds a Markdown study note from Bilibili videos and articles and keeps subtitles and comments beside source links.

SkillExtended access354

A skill that prepares a video, title and cover for five platforms and stops short of the final Publish button.

SkillBroad access276

With douyin-comments, an agent collects comments and replies from one Douyin video into a report and local source records.

SkillExtended access531

Bullshit Detector checks claims from a video, article, tweet or PDF against sources and returns a report scored 0 to 10.

SkillExtended access163

A skill that checks a video for Douyin, Xiaohongshu and Bilibili before publishing and returns a five-level violation report.

SkillExtended access127

README in short

The repository holds `SKILL.md` with a seven-step identity check, the `config.example.md` template, `tools/` scripts for automatic mode and two sample replies. A pitfalls table explains why the reply goes through the notification and not through the field under the video. The README comes in Chinese and English.

README badge

Original SKILL.md

---
name: bilibili-reply
description: 回复 B 站「@我的」评论。当用户说「我@你了」「去B站看看」「回复那条@」等要处理 B 站 @ 通知时使用。流程固定为：核对身份 → 读通知 → 用 video-watch 分析视频 → 回「@我的」界面用回复框发送。回复形式按视频类型决定（教学类详解+时间戳，玩乐类分享情绪）。
---

# B站 @ 回复流程

让 AI 读懂你在 B 站 @ 它的视频，然后在「@我的」界面里回你一条**带可跳转时间戳**的评论。

## 依赖（缺任何一样都跑不起来）

| 依赖 | 说明 |
|---|---|
| **浏览器自动化** ⚠️ | **最关键**。AI 必须能操控一个**已登录 B 站**的浏览器，否则读不到 @ 通知、也发不了评论。推荐 [BrowserSkill](https://github.com/Tencent/BrowserSkill)（`bsk` CLI + 浏览器扩展）。**安装步骤见仓库 README 的「前置条件」一节。** |
| **`video-watch` 技能** | 必需。用它把视频转成带时间戳的文字稿 + 抽帧图片。本技能不重复实现这一步。 |
| **一个 B 站账号** | 建议用不常用的小号；本技能会操作它的评论区。 |

**本技能的示例按 DSH 的 `browser_*` 工具写。** 换环境时把这一步换成你手上的浏览器工具，**流程逻辑不变**——读通知、核对昵称、点回复框这几步是一样的。

**装完先自检**：能在浏览器里打开 `https://message.bilibili.com/#/at` 并读到通知列表，才算前置通了。

## 首次配置

1. 把 `config.example.md` 复制成 `config.md`
2. 按里面说明填入**你的昵称**和**机器人账号昵称 / UID**
3. 确认 `video-watch` 技能已装好（见它的 README）
4. 确认**浏览器自动化可用**（见上「依赖」一节，这是最容易漏的一步）

**为什么要填两个昵称**：本技能的设计前提是「**你用账号 A @ 自己**」——也就是发 @ 的人和执行回复的账号是同一个。所以不能靠 UID 区分，只能靠**昵称核对**。

---

## 第一步：核对身份（最重要，别跳过）

**@ 列表里混着别人的通知。只回目标用户发的，别人的绝对不能回。**

回复前必须走完这个清单（配置见 `config.md`）：

```
① 打开 https://message.bilibili.com/#/at
② 找【最新一条】通知 —— 不要盲取列表第一条，要看时间戳
③ 核对：这条通知显示的【发 @ 人昵称】== config.md 里的 owner_name ?
④ 点这条通知的【评论文本】→ 跳转到视频（URL 带 #reply<评论id>）
⑤ 在视频评论区确认那条评论的作者昵称 == owner_name
⑥ 确认评论里 @ 的账号 == config.md 里的 self_name
⑦ 全部对上 → 才能回复
```

**任何一项对不上 → 停下来问用户，不要擅自回复。**

Report a problem