Building thoughtful software with AI.Notes from the systems behind the work.
Ticking who gets planned works for named staff, but open shifts are grouped by their typed name, so a repeated name or a blank one never reaches the request, and the count on screen can't show it.
Read moreSwitching my chat page to streaming meant errors, drafts and the final answer all had to travel inside the stream, and reading the diff back I think one leak cleanup now cuts too little.
Moving a teams screen from route handlers to server actions deleted two files and kept every permission check, but left the role list in three places, turned refusals into ordinary successful calls, and tested some actions' gates far better than others.
The admin staff list in my scheduling app sent pages with no total, so Next never switched off. The server now counts after the search filter and before slicing, caps the page size, and the screen does the page math; saves also refuse rows with no company, since those could never appear on any page.
A dozen copy-pasted body parsers were replaced with one reader. Broken JSON now gets one clean 400, but the order of its checks turned away body-free deletes and let empty bodies pass the array guards.
The agent's store picker most likely ran its script inside my dashboard on every modal open, so its top-level constants would collide with the copies from the last open; a version wrapped in a function gives each open its own names.
An agent's script-injection patch escaped staff names and chore notes as they leave the database. Reading the code, every save that starts from loaded data writes the escaped text back, and it grows each round. The page it was meant to protect already showed plain text.
Opening the roles endpoint to company admins let them list, edit and delete the one role that wasn't theirs. Each operation names a role differently, so closing it took four checks, and the rename-by-id one is the one I'd have missed.
A new training card pushed my staff schedule's print sheet past one page because nothing counted the title, padding or gaps, and most of my commits answered with thinner margins and a multiplier that shrinks even pages that already fit.
An evening of agent changes kept guessing at sizes for my staff schedule printout, then produced a fixed shrink and a measured shrink for the same element. The merge kept only the measured one, and I can't say whether that was on purpose.
A token, a self-repairing digest and a server deny rule on my staff schedule page all hang off one seed file, and reading the code, a missing seed lets an empty token straight through.
A source-neutral guide to extending a reactive checkout with bounded DOM handoffs, explicit lifecycle state, and safer credential boundaries.