When the same small thing bugs me often enough, I usually end up building a tool for it.
This is where they live. Most of them run entirely on your own machine and don't upload anything.
md2word: Markdown copied out of ChatGPT, Claude or Cursor doesn't paste cleanly into Word, and the math is the worst part. md2word turns it into Word-ready text, a .docx or a PDF, and converts the LaTeX into equations you can still edit in Word. Nothing leaves your computer.
Russian Stress: normal Russian text doesn't show where the stress falls. This Chrome/Edge extension adds the marks to any page, selection or text box, using an offline dictionary of more than 1.1 million word forms. When a word could go either way, like за́мок (castle) and замо́к (lock), it shows both instead of guessing.
youtube-subtitle-downloader: paste a YouTube link, get the subtitles back as clean Markdown with timestamps. Choppy caption lines are merged into paragraphs, and human-made subtitles win over auto-generated ones. It works as a Claude Code skill or a plain CLI.
sub2study goes one step further and turns a video into a bilingual study guide, as a PDF for reading on paper or an iPad and as Markdown for your notes. It strips the rolling repeats out of auto-captions first, keeps a timestamp on every paragraph, and pulls out a vocabulary list with IPA.
russian-learning-app is an earlier take on the same idea: Russian vocabulary taken from real podcasts, reviewed with FSRS spaced repetition.
AttentionFlow: browser history tells you what you opened. It doesn't tell you what scrolled past, what made you stop, or what you read closely. AttentionFlow tracks that on Linux DO and turns your week into a shareable “attention receipt”, with post titles hidden unless you choose to show them. Everything stays in your browser.
GesturePad: a menu bar app for mouse and trackpad gestures. Hold the right button and draw a W to open an app, swipe left to go back, or push the pointer into a corner and scroll to change the volume. You can edit every gesture, and if it misreads a letter, you can record how you write it. It also lets the mouse wheel and the trackpad scroll in different directions, which macOS can't do on its own.
- gemini-tts-deck: a recording-desk style interface for Gemini text-to-speech, from script to voice, style and saved takes.
- agent-atlas: when you run into an unfamiliar agent term, it shows where the term sits next to context, memory and RAG, not just what it means. Read it online.
- minimal-oil-cover: an agent skill that draws minimal oil-painting covers from a photo or a topic.
If something here breaks or you want a feature, open an issue. PRs are welcome too.


