Maybe I missed something about this, but has anyone tried accessing tickets on gitlab.haskell.org using agents? Right now, accessing the content of a ticket using an agent runs into a brick wall, because all the comments are rendered using JavaScript (which the agent won’t run). I suppose I could get an API key for my agent, but I’d rather not go through the fuss.
Getting agents to work there could be a nice boost for those of us using them. I’ve had good experiences with telling Claude “Read github.com/blah/blah/pull/1234 and address all the comments.” or “Review github.com/blah/blah/pull/1234 and tell me what you think.” Of course agents aren’t perfect, but I’ve found this to save me some time.
Agents are a plague to the internet and are causing massive load spikes everywhere, including peoples personally hosted git servers (that affected me too).
I’m not really sure if there’s a way to distinguish between:
agent request initiated by a human and
AI scrapers (for training purposes)
I suppose I could get an API key for my agent, but I’d rather not go through the fuss.
That seems like the right thing to do, since we can somewhat assume requests via that api key were at least at some point initiated by a human. And you can be held responsible if that key is being misused.
The proper way to do this is to teach your agent of choice use the API. The read API is public but (obviously) under a rate limit. Currently the limit is generous enough to make it unlikely that you will hit it with any kind of normal workflow, even an agentic one.
I don’t know what your using but codex automatically used the API for some things. But putting something like Use the REST API path `<path><id>` to fetch gitlab MRs" into your context should probably just work.
There is currently a lot of spam hitting those endpoints so we might throttle them further in the future, but an agent should at most make 4-5 requests there before it can chew on it so I don’t think it will get throttled to a degree where you need an API key soon.
I tried something like this for issues in the past and it worked.
Results from the AI were very hit and miss for me so for now I’m reading those things the old fashioned way. Maybe you will have better luck
