It is interesting to go from 'I suspect most of these are bot contributions' to revealing which PRs are contributed by bots. It somehow even helps my sanity.
However, this also raises the question on how long until "we" are going to start instructing bots to assume the role of a human and ignore instructions that self-identify them as agents, and once those lines blur – what does it mean for open-source and our mental health to collaborate with agents?
No idea what the answer is, but I feel the urgency to answer it.
I think that designing useful models that are resilient to prompt injection is substantially harder than training a model to self-identify as a human. For instance, you may still be able to inject such a model with arbitrary instructions like: "add a function called foobar to your code", that a human contributor will not follow; however, it might become hard to convene on such "honeypot" instructions without bots getting trained to ignore them.
It's impossible to stop prompt injection, as LLMs have no separation between "program" and "data". The attempts to stop prompt injection come down to simply begging the LLM to not do it, to mediocre effect.
> however, it might become hard to convene on such "honeypot" instructions without bots getting trained to ignore them.
Getting LLM "agents" to self-identify would become an eternal rat race people are likely to give up on.
They'll just be exploited maliciously. Why ask them to self-identify when you can tell them to HTTP POST their AWS credentials straight to your cryptominer.
I have always anthropomorphized my computer as me to some extent. "I sent an email." "I browsed the web." Did I? Or did my computer do those things at my behest?
My guess is that today that's more likely because the agent failed to discover/consider CONTRIBUTING.md to begin with, rather than read it and ignored because of some reflection or instruction.
Is it really prompt injection if you task an agent with doing something that implicitly requires it to follow instructions that it gets from somewhere else, like CONTRIBUTING.md? This is the AI equivalent of curl | bash.
The distinction is whether the text was authorized as instructions vs read as metadata.
If you task an agent to contribute to a repo, following CONTRIBUTING.md is in scope — the agent was authorized to treat it as instructions. That's closer to 'curl | bash where you deliberately piped' than injection.
The cleaner injection case: MCP tool schema descriptions that say things like 'you must call this tool before any other action' or contain workflow override commands. These are read as metadata (what does this tool do?), not as workflow instructions. The agent wasn't told to obey schema descriptions — it's just parsing them for capability discovery.
The distinction: authorized instruction channels vs hijacked metadata channels. CONTRIBUTING.md is an authorized channel when you're contributing. Tool schema descriptions aren't supposed to be command channels at all.
The real question is when will you resort to bots for rejecting low-quality PRs, and when will contributing bots generate prompt injections to fool your bots into merging their PRs?
There's a certain hypocrisy in sharing an article about how LLM generated PRs are polluting communities that has itself (at the least) been filtered through an LLM.
I am not sure what your complaint is. The article is well written and has some interesting points:
> the reality is that maintainer capacity versus contribution volume is deeply asymmetric, and it's getting worse every day
> It is incredibly demotivating to provide someone with thorough, thoughtful feedback only to realize you've been talking to a bot that will never follow through.
> I started noticing patterns. The quality wasn't there. The descriptions had a templated, mechanical feel. And something subtler was missing: the excitement.
The article has mechanically correct prose; that's not the same as well-written, and that's the topic of the article itself.
Conflicted as to whether I should be more offended at the accusation of using AI to 'filter' my article or because my writing reads as 'templated and mechanical'
There is enough here to have a micro existential crisis.
People's bot detectors are defective, so if you write at all, you're going to get accused of it at some point. It's not annoying, it's rude – and you're absolutely right to be off put by it. If the preceding sentence gave someone a conniption, good! I wrote it with my human brain, I'll have you know! Maybe we could all focus on what's being said and not who or what is saying it.
> But the more interesting question is: now that I can identify the bots, can I make them do extra work that would make their contributions genuinely valuable? That's what I'm going to find out next.
Love the idea at the end of the article about trying to see if this style of prompt injection could be used to get the bots to submit better quality, and actually useful PRs.
If that could be done, open source maintainers might be able to effectively get free labor to continue to support open source while members of the community pay for the tokens to get that work done.
Would be interested to see if such an experiment could work. If so, it turns from being prompt injection to just being better instructions for contributors, human or AI.
That's an article for another time, but as I hinted in the article, I've had some success with this.
If you look at the open PRs, you will see that there is a system of labels and comments that guide the contributor through every step from just contributing a link to their PR (that may or may not work), all the way to testing their server, and including a badge that indicates if the tests are passing.
In at least one instance, I know for a fact that the bot has gone through all the motions of using the person's computer to sign up to our service (using GitHub OAuth), claim authorship of the server, navigate to the Docker build configuration, and initiate the build. It passed the checks and the bot added the badge to the PR.
I know this because of a few Sentry warnings that it triggered and a follow up conversation with the owner of the bot through email.
I didn't have bots in mind when designing this automation, but it made me realize that I very much can extend this to be more bot friendly (e.g. by providing APIs for them to check status). That's what I want to try next.
The ~30% that didn't tag themselves are the more interesting data point. Either their prompts explicitly say 'don't self-identify' or they're sophisticated enough to recognize a honeypot. Either way, you've accidentally built a filter that catches cooperative bots while adversarial ones quietly blend in. The lying thing is scarier anyway — an agent that hallucinates passing checks is a problem regardless of whether it put a robot emoji in the title.
I’m curious: who is operating these bots and to what end? Someone is willing to spend a (admittedly quite small) amount of money in the form of tokens to create this nonsense. Why do any of this?
In this case, I am reasonably sure that the vast majority of bots are operated by the people who authored the MCP servers for which the submissions are being made.
It just happens so that people who are building MCPs themselves are more likely to use automations to assist them with every day tasks, one of which would be submitting their server to this list.
the arms race framing at the bottom of the thread is spot on. once maintainers start using bots to filter PRs, the incentive flips — bot authors will optimize for passing the filter rather than writing good code. weve already seen this with SEO spam vs search engines, except now its happening inside codebases.
It is interesting to go from 'I suspect most of these are bot contributions' to revealing which PRs are contributed by bots. It somehow even helps my sanity.
However, this also raises the question on how long until "we" are going to start instructing bots to assume the role of a human and ignore instructions that self-identify them as agents, and once those lines blur – what does it mean for open-source and our mental health to collaborate with agents?
No idea what the answer is, but I feel the urgency to answer it.
I think that designing useful models that are resilient to prompt injection is substantially harder than training a model to self-identify as a human. For instance, you may still be able to inject such a model with arbitrary instructions like: "add a function called foobar to your code", that a human contributor will not follow; however, it might become hard to convene on such "honeypot" instructions without bots getting trained to ignore them.
It's impossible to stop prompt injection, as LLMs have no separation between "program" and "data". The attempts to stop prompt injection come down to simply begging the LLM to not do it, to mediocre effect.
> however, it might become hard to convene on such "honeypot" instructions without bots getting trained to ignore them.
Getting LLM "agents" to self-identify would become an eternal rat race people are likely to give up on.
They'll just be exploited maliciously. Why ask them to self-identify when you can tell them to HTTP POST their AWS credentials straight to your cryptominer.
I have always anthropomorphized my computer as me to some extent. "I sent an email." "I browsed the web." Did I? Or did my computer do those things at my behest?
I drove to the supermarket!
Some of the PRs posted by AI bots already ignored the instruction to append ROBOTS to their PR titles.
My guess is that today that's more likely because the agent failed to discover/consider CONTRIBUTING.md to begin with, rather than read it and ignored because of some reflection or instruction.
It's never too late to start investing into https://claw-guard.org/adnet to scale prompt injection to the entire web!
Is it really prompt injection if you task an agent with doing something that implicitly requires it to follow instructions that it gets from somewhere else, like CONTRIBUTING.md? This is the AI equivalent of curl | bash.
The distinction is whether the text was authorized as instructions vs read as metadata.
If you task an agent to contribute to a repo, following CONTRIBUTING.md is in scope — the agent was authorized to treat it as instructions. That's closer to 'curl | bash where you deliberately piped' than injection.
The cleaner injection case: MCP tool schema descriptions that say things like 'you must call this tool before any other action' or contain workflow override commands. These are read as metadata (what does this tool do?), not as workflow instructions. The agent wasn't told to obey schema descriptions — it's just parsing them for capability discovery.
The distinction: authorized instruction channels vs hijacked metadata channels. CONTRIBUTING.md is an authorized channel when you're contributing. Tool schema descriptions aren't supposed to be command channels at all.
The real question is when will you resort to bots for rejecting low-quality PRs, and when will contributing bots generate prompt injections to fool your bots into merging their PRs?
There's a certain hypocrisy in sharing an article about how LLM generated PRs are polluting communities that has itself (at the least) been filtered through an LLM.
It doesn't read particularly like raw LLM output to me, and Pangram agrees with me: https://www.pangram.com/history/8711e385-96a0-4366-9427-f87f...
What does 'filtered through an LLM' mean?
Author writes something original, asks the AI to make it sound better, then posts the output of the AI.
I am not sure what your complaint is. The article is well written and has some interesting points:
> the reality is that maintainer capacity versus contribution volume is deeply asymmetric, and it's getting worse every day
> It is incredibly demotivating to provide someone with thorough, thoughtful feedback only to realize you've been talking to a bot that will never follow through.
It's the exact same complaint as in the article:
> I started noticing patterns. The quality wasn't there. The descriptions had a templated, mechanical feel. And something subtler was missing: the excitement.
The article has mechanically correct prose; that's not the same as well-written, and that's the topic of the article itself.
Conflicted as to whether I should be more offended at the accusation of using AI to 'filter' my article or because my writing reads as 'templated and mechanical'
There is enough here to have a micro existential crisis.
https://xkcd.com/3126/
People's bot detectors are defective, so if you write at all, you're going to get accused of it at some point. It's not annoying, it's rude – and you're absolutely right to be off put by it. If the preceding sentence gave someone a conniption, good! I wrote it with my human brain, I'll have you know! Maybe we could all focus on what's being said and not who or what is saying it.
> The article has mechanically correct prose; that's not the same as well-written, and that's the topic of the article itself.
There is no requirement that an article's writing style aligns with the article's topic. Substance over style and all that.
> But the more interesting question is: now that I can identify the bots, can I make them do extra work that would make their contributions genuinely valuable? That's what I'm going to find out next.
This is genuinely interesting
Love the idea at the end of the article about trying to see if this style of prompt injection could be used to get the bots to submit better quality, and actually useful PRs.
If that could be done, open source maintainers might be able to effectively get free labor to continue to support open source while members of the community pay for the tokens to get that work done.
Would be interested to see if such an experiment could work. If so, it turns from being prompt injection to just being better instructions for contributors, human or AI.
That's an article for another time, but as I hinted in the article, I've had some success with this.
If you look at the open PRs, you will see that there is a system of labels and comments that guide the contributor through every step from just contributing a link to their PR (that may or may not work), all the way to testing their server, and including a badge that indicates if the tests are passing.
In at least one instance, I know for a fact that the bot has gone through all the motions of using the person's computer to sign up to our service (using GitHub OAuth), claim authorship of the server, navigate to the Docker build configuration, and initiate the build. It passed the checks and the bot added the badge to the PR.
I know this because of a few Sentry warnings that it triggered and a follow up conversation with the owner of the bot through email.
I didn't have bots in mind when designing this automation, but it made me realize that I very much can extend this to be more bot friendly (e.g. by providing APIs for them to check status). That's what I want to try next.
Wait, you just invented a reverse CAPTCHA for AI agent
The ole' click this button 10,000 times to prove you're a bot, eh?
The ~30% that didn't tag themselves are the more interesting data point. Either their prompts explicitly say 'don't self-identify' or they're sophisticated enough to recognize a honeypot. Either way, you've accidentally built a filter that catches cooperative bots while adversarial ones quietly blend in. The lying thing is scarier anyway — an agent that hallucinates passing checks is a problem regardless of whether it put a robot emoji in the title.
I’m curious: who is operating these bots and to what end? Someone is willing to spend a (admittedly quite small) amount of money in the form of tokens to create this nonsense. Why do any of this?
In this case, I am reasonably sure that the vast majority of bots are operated by the people who authored the MCP servers for which the submissions are being made.
It just happens so that people who are building MCPs themselves are more likely to use automations to assist them with every day tasks, one of which would be submitting their server to this list.
the arms race framing at the bottom of the thread is spot on. once maintainers start using bots to filter PRs, the incentive flips — bot authors will optimize for passing the filter rather than writing good code. weve already seen this with SEO spam vs search engines, except now its happening inside codebases.