jhogendorn

- Got https://beachcomber.sh pretty much stable. Next stage is to propose to various upstreams its worth integrating. - Custom firmware for some ikea symfonisk dials because the oem firmware on them has some pretty bad bugs. Added features like hold and turn. Getting nice smooth dial behaviour over zigbee etc is surprisingly tricky - Built a skill evaluator tool that runs a skill through test suites and then tweaks the skill context and runs again. Its been pretty effective to be honest, almost all skills you do the first version is laughable compared to the one you get after this automated self improvement. - A robust tmux bridge interface for claude to hook into, and then a director layer on top of that for agent orchestration tooling - a stenographer skill that on the fly ripgrep and builds a rag on your on disk conversation history as a form of memory. Pretty effective. - I have just started a tool that brokers woodpecker ci to openbao/vault to give a gitlab like integration for controlled secrets injection for ci. - Been beating my head against a camera tool for a while now, finally making headway. Many ptz cameras dont support fov move, which nvrs need for ml object detection and tracking. They just have a super clunky continuous move and stop. So my tool characterises the camera with cv tools and calibrates movement curves to produce a data file that can be used by my onvif proxy to emulate the more advanced move commands. - Various helper tools for fusion, like csv based parameterised export, and compliant magnet insert generators. - A pipeline that consumes my content backlog, ie instagram saves, reddit saves, hn faves, etc and analyses them with local models and various algorithms steps to categorise and intuit why it was saved and what the key information is and what category it fits into for future reference etc. - A map of my city that shows live river height data with flood map overlays, contour data, predicted overland flow etc. flooding is a regular concern but theres no great resource to know whats going on. I have about 60gb of public datasets it works with. - A package manager for kicad library symbols and footprints, datasheets - skills for kicad so claude can reasonably interpret the schematic and advise on problems, check against datasheets etc. surprisingly effective. - A gcode controlled expansion board for the Carvera Air that gives you 8+8 channels of control for extraction, air assist, vacuum table, timelapse camera, etc. you only have 1 pwm pin so the protocol encodes over that. - A novel exploration interface for vitamins that renders them in a network graph, showing relationships. When you select one it rearranges around it into a kind of valance orbit style so you can explore chains of effect. Turns out, lots and lots of things relate to magnesium. - A comprehensive usb c pd board with 4s battery management. 3a or 8a depending on version. Trying to do proper pd in is nontrivial so this is a drop in solve. - A new brain pcb for Kinesis Advantage Pro keyboards to give modern firmware, bluetooth etc. - Repacked my rack UPS battery with LiFePo cells, and built an induction/resistive series battery balancer pcb for it. - Playing around with a new debug header/connector concept thats tiny footprint and zero cost to add.

The hard part is getting things over the line, publishing and seeing if theres interest. A thing can be largely done but theres a lot of detail work polishing it up so its public ready.

I probably would not have bothered to allocate time to this pre ai, the juice wasnt really worth the squeeze. But I approached it with an initial amusing naivete about it being 'super simple'. As is almost always the case with software theres a reasonable amount of hidden complexity. But I have been using it as a sort of learning proving ground for how to work with agentic development. For example I got to a point where claude wouldnt implement properly and would argue with me about changes because it would read the current/old docs in the repo and get confused about reality. So right now I'm experimenting with 'canonical specs' that can only be changed modally with gates and a defined cascade from canon, to code, to docs in that order. Otherwise you end up in a weird thrash about the docs and the code disagree and which one will the agent decide to change for consistency?

Anyway, its been interesting and its v0.6 and at a point where Ive not hit a sharp edge dogfooding for a while and some beta testers would be valuable. Right now you have to manually wire it into your stack, once some others have kicked the tyres hard enough I will make some pr's to the popular tools to consider integrating it.