I ran a cost estimate on the project I describe in https://simonwillison.net/2025/Oct/23/claude-code-for-web-vi... - which was covered by my Claude Max account, but I dug through the JSONL log files for that session to try and estimate the cost if I had been using the API.
Once you switch to ijson it will not save any memory, no, because ijson essentially uses zero memory for the parsing. You're just left with the in-memory representation.
1. there are hallucinatory descriptions in the Readme (make test), and also in the code, such as the rate limit set at line 158, which is the wrong number
2. all commits are done on github webui, checking the signature confirms this
3. too verbose function names and a 2000 line python file
I don't have a complaint about ai, but the code quality clearly needs improvement, the license only lists a few common examples, the thresholds for detection seem to be set randomly, _get_stargazers_graphql the entire function is commented out and performs no action, it says "Currently bypassed by get_ stargazers", did you generate the code without even reading through it?
Bad code like this gets over 100stars, it seems like you're doing a satirical fake-star performance art.
I checked your past submissions and yes, they are also ai generated.
I know it's the age of ai, but one should do a little checking oneself before posting ai generated content, right? Or at least one should know how to use git and write meaningful commit messages?
It's a project I'm making purely for myself and I like to share what I make - sorry I didn't put up most effort in the commit messages, will not do that again.
Well I initially planned to use GraphQL and started to implement it, but switched to REST for now as it's still not fully complete, just to keep things simpler while I iterate and the fact that it's not required currently. I’ll bring GraphQL back once I’ve got key cycling in place and things are more stable. As for the rate limit, I’ve been tweaking things manually to avoid hitting it constantly which I did to an extent—that’s actually why I want to add key rotation... and I am allowed to leave comments for myself for a work in progress no? or does everything have to be perfect from day one?
You would assume if it was pure ai generated it would have the correct rate limit in the comments and the code .... but honestly I don't care and yeah I ran the read me through GPT to 'prettify it'. Arrest me.
`ecapture` has been around for a while and do a lot of great stuff and a lot of functionality overlaps.
Our aim is to make Qtap extensible and via a plugin system. We have http1/2 streaming capabilities and a plugin engine to run these in what we call a stack. Our goal is to add more protocols, like gRPC in the near future.
We have a few example plugins that do things like report request/response's and push access information to standard out in a console or structure log format. Our Pro version has a few more plugins like the ability to report errors (eg. an AI agent is getting HTTP 429 errors). These can be pushed to a service or log aggregator.
To summarize, we do a lot of the same things that ecapture does. We'd like to be less of a tool and more of a "always running" that ops, opsec, and devs use to answer tough questions. We look forward to open sourcing more of plugins as they mature!
What about performance overheads? I think the reason a lot of people write C is that they have direct control over the generated assembly to maximise performance.
Even direct control over the generated assembly (C does not provide this; only writing assembler does, and only if you don't do things like LTO) is not sufficient.
Modern CPUs do all sorts of weird things. Assembly instructions can be executed out of order. Your conditional jump instruction can be speculatively executed before the condition's truth is known. Fetches from main memory can be reordered.
Even more wildly, copying the contents of one register to another is often a no-op. Yes, that's right; the following code:
mov edx, eax
... does next to nothing on some modern CPUs. All it does sometimes is set an internal note to the effect of "later references to edx should read/write eax instead", until that note is cleared by some other operation.
You can write your assembler with the best of intentions as to how it should behave, only to discover that the CPU does things entirely differently. You still end up getting the same observable result out of it, but any timing and order of operation guarantees went out the window decades ago.
There is no point in asking such questions, the model doesn't know what it is on its own, and you could get many different answers if you repeat it a few more times.
Looks like a typical mitm attack, which confuses me a bit, don't the developers use something like tls or dtls to protect their communications?
The most recent game I analysed was helldivers 2, which uses dtls. i would have thought that would be fairly common knowledge.
That's the issue - they don't!
But even games like Helldivers 2 have had silly vulnerabilities (just not RCE) - see https://helldivers.io/freesupercredits for some examples.
