Bluesky reply bot checker

Simon Willison released a tool called Bluesky reply bot checker on September 27, 2026. By his own account, the motivation was direct: automated reply bots are rampant on Twitter, and as someone with a sizable following, every post he makes attracts a large number of mindless automated replies; and this kind of bot has already begun appearing on Bluesky.

So he had Opus 5.5 write this tool in a vibe-coding way. The tool's positioning is not to intercept or ban, but to check any Bluesky profile for evidence that it might be a reply bot—turning "does this reply look like a bot" from intuition into a checkable basis for judgment.

According to the source description, the tool looks for three types of signals. The first type is a timing anomaly: an account posts a reply within seconds after another account posts. The second type is a content anomaly: an account never posts its own content, images, or links, but instead constantly replies to other users with more followers. The third type carries obvious personal emotion—the tool looks for question marks, because the author is especially annoyed by reply bots that "waste his time answering a question no one actually asked."

These three types of signals point to the same behavioral pattern: not producing content, only reacting quickly to high-exposure accounts, and using question-style phrasing to induce interaction. It should be noted that the source only describes which signals the tool looks for; it does not explain how these signals are combined into a final judgment, nor does it provide any detection accuracy or false-positive rate data.

The author points out a key difference in the article: unlike Twitter, Bluesky still provides a freely usable and valuable API. He also emphasizes that the lack of a usable API does not slow down the bots themselves, but it makes investigating them more difficult.

This sentence splits the problem into two layers. Whether bots are rampant has nothing to do with whether a platform opens its API; but whether researchers and ordinary users can investigate a suspicious account depends heavily on whether the API is available. The reason the tool can exist is the premise that data can still be retrieved—this is the only structural condition explicitly given in the source.

What can be confirmed is: the tool was written by Simon Willison using Opus 5.5 in a vibe-coding way, published on September 27, 2026, and its purpose is to check Bluesky profiles for reply bot evidence and to look for the three types of signals above.

What cannot be confirmed is equally clear. The source is a blog post by the author himself, and it does not provide the tool's detection accuracy or false-positive rate; it does not state whether the tool is open source, publicly accessible, or actually deployed; it does not give the time or code size required for Opus 5.5 to generate the tool; and it does not provide quantitative data on the number or proportion of reply bots on Bluesky. Therefore, the tool's actual effectiveness at identifying reply bots, the extent to which Opus 5.5 completed it independently, how much the author participated, and whether the tool will be adopted by Bluesky officially or have any actual impact are all currently impossible to judge.

Putting this matter back at its own scale: it is not a platform governance plan, nor a bot census report, but a small tool that one person quickly made with an off-the-shelf model after being annoyed by reply bots.

Its value is not that it "solved the bot problem," but that it demonstrates a path—when a platform is still willing to open its API, an individual can, at extremely low cost, turn "I think this is a bot" into "I can go check." As for how far this path can go, it depends on whether the tool itself is reliable, and the source does not answer that.