Redoing the work as smaller PRs might help with readability, but then you get the opposite problem: it becomes hard to hold all the PRs in your head at once and keep track of the overall purpose of the change (at least for me).
IMO the real solution is figuring out which subset of changes actually needs human review and focusing attention there. And even then, not necessarily through diffs. For larger agent-generated changes, more useful review artifacts may be things like design decisions or risky areas that were changed.
We use a mixture of LLMs (e.g. to identify the important variables to trace), language servers (to understand where a variable is originally defined and where it's used).
We may go back to code visualization for editors, but after testing it with a lot of engineers, we found that it was too confusing for code reviews. The main issue was just how alien the interface is!
> 1. Allow me to step through the code execution paths that have been modified in the pull request, based on the tests that have been modified.
Not sure if I fully grasp this! We tried to kind of do this in previous iterations (show call graphs all at once) and it gets messy very fast. Could you elaborate on this point in particular?
Sure - imagine my PR adds one new test which test one new function.
Starting from the test, allow me to step through the program execution, just like a debugger, to observe variables, surrounding code, and the complete file.
If you read only the covered lines of code in a linear way, you'd miss the refactoring opportunities because you aren't looking at the rest of the file.
Yes! I personally iterate without understanding the code too much (since it'll change), and fully assess what Claude Code (in my case) has done after I finished a piece of work.
Not sure if I fully grasp what you mean by dog whistling, but at the end of the day, like another commenter said, Haystack is also pretty helpful for when you're done experimenting with a piece of work and need to see what an AI has generated.
I think this is a valid part of the "crafting PR" skill that's under appreciated, and part of the goal of Haystack here is to make that part of PR craft effortless.
Sorry for the confusion! Missing browser navigation is the problem. The virtual back buttons you put in the top left are working as I'd expect browser nav to do. I keep trying to go back, it would feel so natural.
Redoing the work as smaller PRs might help with readability, but then you get the opposite problem: it becomes hard to hold all the PRs in your head at once and keep track of the overall purpose of the change (at least for me).
IMO the real solution is figuring out which subset of changes actually needs human review and focusing attention there. And even then, not necessarily through diffs. For larger agent-generated changes, more useful review artifacts may be things like design decisions or risky areas that were changed.
reply