Skip to content
Kokosa's Intensive Care Unit
Kokosa's Blog

Stuffing an RSS Subscription into the Bot

- 3 min read - 600 words
On this page

The Current Tools Are Too Hard to Use… So…

Section titled “The Current Tools Are Too Hard to Use… So…”

The forwarding bot I wrote earlier was originally just for receiving private messages. Because I needed ad blocking, I added LLM moderation. To save a bit of token, I casually added allowlists and blocklists. To prevent false bans, I added an appeal system. Then, with nothing better to do, I added i18n… oh crap, how did this get to nearly 2,000 lines of code? The RSS tool I use also happens to be pretty clunky, so why not stuff an RSS reader in there too? kfb, one day you’re going to become a bot with way too many features…

My requirements are very simple: when a friend’s (see the friend links) blog updates, send me a Telegram message. No need to make it into some full-blown feed reader (because I think current reader designs are all pretty ugly). A few commands to add, remove, update, and list RSS feeds, plus basic send/receive — that’s enough. So:

/rss_add              # add
/rss_list             # list
/rss_remove           # remove
/rss_refresh          # refresh
/rss_title            # rename
RSS Bot Command List
RSS Bot Command List

The Mysterious World of RSS Generation

Section titled “The Mysterious World of RSS Generation”

Here’s a relatively ideal item:

<item>
  <title>Article Title</title>
  <link>https://example.com/post</link>
  <guid>https://example.com/post</guid>
  <description>Article summary</description>
  <pubDate>Fri, 08 May 2026 00:00:00 GMT</pubDate>
</item>

However, every site generates things slightly differently. Some don’t have content:encoded (like Dongye’s blog doesn’t have one), some use the link as guid, and some use some mysterious ID as guid.

For example, here’s mine:

<item>
<title>test-2</title>
<link>https://kokosa.icu/zh-cn/blog/test-2/</link>
<guid isPermaLink="true">https://kokosa.icu/zh-cn/blog/test-2/</guid>
<description>This is a test article</description>
<pubDate>Fri, 08 May 2026 00:00:00 GMT</pubDate>
<content:encoded>
<p>114514 characters omitted...</p>
</content:encoded>
<category>Test</category>
</item>

I’ve always felt that stuffing the full article into RSS is incredibly stupid, so I’ve already changed this.

Pushing Old Posts — Enjoy Getting Flooded ()

Section titled “Pushing Old Posts — Enjoy Getting Flooded ()”

When adding an RSS feed for the first time, you definitely can’t just push all the articles at once — it’s a message bot after all. Some blog RSS feeds contain dozens of articles (and I’ve already read them all). Adding a subscription and flooding the whole screen would be a nightmare. So when first adding a subscription, it fetches the feed once and marks all current articles as read, but doesn’t send them. After that, scheduled refreshes only send notifications when new items are discovered:

[RSS] Kokosa's Blog


Stuffing an RSS Subscription into the Bot
https://kokosa.icu/zh-cn/blog/2-tgbot-rss.md


Summary: Added RSS subscription notifications to the forwarding bot, with scheduled refreshes, deduplication, and AI summaries.

Cloudflare Cron

Section titled “Cloudflare Cron”

A Worker isn’t a persistent service, so to do scheduled refreshes you need to write this thing in wrangler.toml:

[triggers]
crons = ["*/30 * * * *"]

It checks every 30 minutes. For easier testing, I added a manual command:

/rss_refresh

Then I realized that manual refreshes might collide with Cron, so it needs a KV lock. Otherwise two refresh tasks might both find “this article hasn’t been sent yet” and send it twice.

Notification after manually refreshing RSS
Notification after manually refreshing RSS

AI-Generated Article Summaries

Section titled “AI-Generated Article Summaries”

Since Gemini was already integrated, how could a notification contain nothing but a link?! So I made it summarize the article too. The content moderation timeout was previously set to only 4 seconds, because moderation just needs to output SAFE or UNSAFE — whereas summaries have a lot more to think about. So summaries get a separate 20-second timeout. If it fails, the worst that happens is:

Summary: AI summary unavailable.

The link still gets sent as usual.

Afterword

Section titled “Afterword”

After vibing through this at lightning speed, I realized: couldn’t I have used something like OpenClaw or Hermes to do this for me? (kfb, rise up… surpassing the lobster is only a matter of time…)

Tags: