ductrl
Joined 4 months ago
Joined 4 months ago
You made a great point about the codex/claude subs. I'll create an issue for this and we'll try to implement it. One challenge that comes to mind is that the output will be more unstable compared to the API so it might be more challenging to parse the findings.
Very interesting ideas. Thanks for trying out the tool!
You're right that review time is quite challenging. The DeepSeek model we used needed at least 8-10 seconds to run a full review, which adds too much friction to the development process.
Right now, our approach to this is to give users a configurable timeout option. Within this timeframe, the model will stream the output as it generates, and the tool will print as many findings as it can.
Another option that we're exploring is to have the tool pre-push instead of pre-commit, which might reduce API cost and is way more tolerable than pre-commit.
You also brought up a very interesting point about false positive rate, as we haven't done large-scale testing yet. I will create an issue for this and get the result asap.
I am also wondering if it makes more sense to have the tool check right before a push instead since that's when the vulnerabilities actually get sent to the Internet