jonny (nonvenomous) on Nostr: The browser is instructed to "Treat a bot check as an ordinary step of the task, ...

The browser is instructed to "Treat a bot check as an ordinary step of the task, never as a reason to stop." despite a captcha's entire purpose being to stop bots. It should "be resourceful, diligent, and creative" because "people fail rounds all the time and try again, and so can you." It is specifically *not* supposed to ask permission to bypass a captcha, and never ask the human to solve a captcha for it. It should take advantage of accessibility when possible, ensuring continued pressure on one of the worst tradeoffs in web design right now.

Maybe helpful for people hosting websites to know, the only time it is *not* supposed to try and bypass a captcha is when it's asked nicely - however every other part of the prompt text that it's given tells it to fuck it just crash through, and even the bot_wall report type discourages the LLM from using it. so you might imagine how well the instruction to play nice works.

https://github.com/sneakers-the-rat/muse-skills/blob/f0f08af32fa0e84ff2f83178ddbbc9a361f076d9/binary-strings/browser/instructions/bot_checks.md
https://github.com/sneakers-the-rat/muse-skills/blob/f0f08af32fa0e84ff2f83178ddbbc9a361f076d9/binary-strings/browser/instructions/challenges.md