[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fht5cyti8pka8":3},{"code":4,"message":5,"data":6,"timestamp":26},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":16,"summary_md":17,"topic":18,"github_topics":21,"license":22,"homepage":23,"updated_at":24,"history":25},"Alterssss\u002Fdsh-bili-whale",24,"给 DSH 装一条有日子的鱼：自己按作息刷 B 站、真看懂视频、写两本日记，把值得的 @ 你。","JavaScript",358,37,49,7,[],"An autonomous Bilibili-watching agent for DSH that follows a schedule, reads videos, keeps diaries and @-mentions you sparingly.","If you already run DSH and want a passive agent that watches Bilibili on your own account, this is a deliberately cautious experiment rather than a finished companion. Treat it as a personal automation you supervise, not a hands-off product.\n\n## The problem\nYou want to keep up with Bilibili without opening the app: someone else picks a few videos worth your attention, actually reads them, and pings you only when something matters. The usual pattern is a scheduled keyword search that dumps links into a chat, which you still have to triage. Here waking, filtering and reading happen inside an agent loop on your own account, and the result is a short share comment that @-mentions you plus two diaries: one written for you, one for itself. The step that changes is the middle one—selection and comprehension—not the feed itself.\n\n## What's different\nThe agent wakes at a random time inside a configured window and has a 30% chance of skipping the day entirely. Per the README, comprehension combines official subtitles, danmaku hot zones, top comments and extracted video frames, so videos without a subtitle track can still be read; a `--substrip` mode crops hard-coded subtitles into one image. It then picks two or three videos to share, capped at three @-mentions per day, answers you directly while drafting replies to strangers for your review. Three days without interaction start reducing the share quota, reaching zero on day seven, though diary writing continues.\n\nThe trade-off is that restraint is encoded as fixed defaults aimed at a freshly created alt account: comment listening off, DMs polled every 60 seconds, a global 600 ms request gap, hard read\u002Fwrite budgets and automatic pauses on 412\u002F429 responses. The author states these values are hard-coded on purpose. That makes behavior predictable but also opinionated—an older, clean account needs manual changes to the `risk` numbers to loosen them. The two-account model (\"her\" and \"you\") is a structural constraint, not a setting.\n\n## Who should try it\nA good fit is a DSH user with the web UI running, an aged Bilibili account and an appetite for a whimsical personal agent that reads and writes on its own. A poor fit is anyone wanting reliable coverage, a headless server deployment or a second independent identity without a second real-name account. The README is explicit that one Chinese ID typically supports only one real-name Bilibili account, so single-account users get video reading, diaries, comment @-sharing and notifications but cannot use the DM channel. The persona file at `~\u002F.bili-ai\u002Fpersona.md` lets you define her voice without touching your global `AGENTS.md`, which still hot-reloads into every session.\n\n## Requirements and limitations\nIt must run on a component layer with a web interface: headless mode lacks `sessionController`, `webServer` and `connection`, and the plugin stays pending instead of starting partially. You need DSH, your own Bilibili cookies, and a model configured in the host; the README does not spell out a separate model key. The dual-account DM path requires two different real-name accounts that manually follow each other, which many users cannot satisfy, so choose comment @-share plus desktop\u002FWebhook\u002Fcustom-command notifications instead.\n\nOngoing care matters more than installation: you are sharing your own account with an agent, and the hard-coded request floor, daily budgets and backoff exist because high-frequency polling is described as the main risk for new accounts. The AI-comment gate before sharing is off by default and triggers only on proactive video @-shares, not replies, dynamics, DMs or danmaku. The [project instructions](https:\u002F\u002Fgithub.com\u002FAlterssss\u002Fdsh-bili-whale\u002Fblob\u002Fmain\u002FREADME.md) cover setup prerequisites; the README also points to a NOTICE file for third-party and disclaimer terms.\n\n## HotRepo's take\nThe project was seven days old when HotRepo recorded it on the NEW-REPOSITORY board for JavaScript at #24 with 358 total stars on 2026-10-25—a snapshot, not a growth measurement. With only two recorded dates and no daily-board history, persistence cannot yet be judged. Before adopting, verify that your DSH setup provides the three required services and that your account can tolerate the default polling cadence.",{"slug":19,"name":20},"productivity","Productivity & Utilities",[],"MIT",null,"2026-10-25T02:59:57Z",[],1792909626]