This was actually really interesting to read. I think the coolest part isn’t even the tools themselves, but how you keep changing your workflow when you find a better way of working. A lot of people just stick with whatever setup they started with, so seeing you experiment and rebuild the process is pretty refreshing. Really enjoyed this one!
Using the issue tracker as a prompt repository is a smart idea. The stacked PR workflow also seems much easier to manage at scale.
netproxy
The issue tracker as a prompt repository is a framing I want to borrow. One question though: the stacked PR batch still needed several rounds of review feedback and CI fixes, so is the real win in parallelizing the mechanical work while the judgment-heavy review part stays serial? The CPU and memory cost of dozens of subagents also sounds worth tracking over time. Thanks for writing it up in detail.