A lot of us here are turning into AI vampires now. AI work does not stop at 9pm. The screen stays on, the terminal feeds paragraph upon paragraph of white text, and your eyes pay for it. I built Shade_ because I am dyslexic and I grade footage, which is the worst possible combination of reasons to let Night Shift warm your whole display. Night Shift cannot leave your grading window alone. Shade_ can. It warms everything you are not actively grading or editing, and cuts a live Passthrough hole over the windows that need honest colour. The hole follows the window. Resize it, the hole resizes. The honest part: it is a source-over overlay, so it adds warmth, it does not subtract pixels. That one macOS constraint shaped the whole product, which is also why the experimental pixel-transform engine is still dev-gated, and why the tool is free. ———————————————————————————— How I actually made it //Using ARI-OS I did not sit and type this app out. I directed it, Claude executed. The shape of the build: 1. One brainstorming session to define the outcome. 2. Three parallel Claude agents to build it. 3. Five simultaneous Claude agents and one Codex agent to debug it and hunt for any errors. 4. One round of manual testing, by me. 5. One more agent run to ship the install page, the open-source GitHub repo, and the Pushing Squares product page. The skill was never in the typing. It was in knowing exactly what honest colour, a click-through overlay, and a dyslexia reading tint needed to be before a single line existed. ———————————————————————————— Three doors Try it, free: https://www.pushingsquares.com/shade The code, open source: https://github.com/PUSHINGSQUARES/Shade_ The full deep-dive on how and why: https://aris-space.com/documents/tools-and-plugins/shade //A<3

I'm in over my head and that's fine by me. Wondering how others are managing context windows for big agent projects. Almost like every client should have their own $100 max plan I'm thinking.