I spent the day dogfooding aidd against itself, and it did not come out clean.

The audit cycle ran against aidd’s own repository. One problem involved the Coordinator’s fleet-level maintenance suggestions: coding runs were treating that advice as work to implement. I changed the selection rule so coding mode takes feature-backed backlog work, while Coordinator priority remains advice. Audit maintenance and completion markers needed fixes too. The run output gave me enough detail to work out what had gone wrong and change the rules before the next run.

I also spent some time on the web panel: the Coordinator layout, recipe view toggles, dashboard organization, artifact details, and report forms. There were accessibility fixes and problems with live run output to sort out as well.

Diagram: the dogfooding loop. Running aidd against its own repository exposes a flaw, its own run output shows why, the rule is corrected, and the next run uses the corrected rule run aidd on its own repo it gets it wrong a real flaw the run shows why its own output rule fixed same day the next run uses the corrected rule the four rule areas this pass corrected selecting work Director suggestions audit maintenance completion marks
Diagram of the four areas where these runs exposed rules that needed fixing. It uses the later name Director; in May, this was the Coordinator.

This is what I’d hoped would happen when I started using aidd on itself. It still gets things wrong, but I can follow the run afterward and usually see why.