What have I done: my open source over the last few months

Usually I write here about other people's releases. But over the last few months I've accumulated several of my own open source micro-projects, and I've collected them into one post. Everything (almost) under MIT, everything free, you can use it, or you can fork it and take it apart.


📕 Texts and notes

edit.computer - a markdown editor that works entirely in the browser and writes directly to your files: to a local folder or to any server via SSH. No account, no installation: an Obsidian vault or a folder with posts for Jekyll opens as is. The connection is encrypted end-to-end from browser to server, so the intermediate server only sees ciphertext, without keys or files.
→ edit.computer


🔄 Context for AI

Copy2LLM - a layer between the site and any AI. The reader puts pages and selected fragments into a local "context basket", sees how many tokens it has collected, and takes it to ChatGPT, Claude, Perplexity, or to their own address. Everything is in the browser, no servers in the chain.
→ copy.computer


👨‍💻 For maintainers

pullguard - a GitHub Action that estimates the "review cost" of a pull request on a scale from 0 to 100. It doesn't try to guess whether a patch was written by a human or AI: the signal is noisy and gives the maintainer nothing. Instead, it calculates how much time the review will take: almost empty changes, refactorings without tests, descriptions that diverge from the diff. It works with your OpenAI or Anthropic key and, according to a configured policy, comments, adds labels, or closes the pull request.
→ arbatov.dev/pullguard


🤖 For agents

herd - a small coordinator for multi-agent work for the terminal herdr. The "head" agent Claude Code takes tasks, distributes them to executors (Claude, Codex, Gemini - whoever you say) in separate tabs, restarts failed ones, and collects reports. Three files, no daemon.
→ github.com/vladzima/herd

prompt-comments - an Agent Skill that prevents files claude.md and agents.md from growing. After six months, no one remembers why a particular line appeared in the file, and it's scary to touch it. The skill forces the agent to record the "why" right at the moment when a rule is added: what broke and how it was fixed. The idea is from the work "Why Does CLAUDE.md Keep Growing?".
→ github.com/vladzima/prompt-comments


🧰 Little things for life

solar-status - a summary of solar activity for your city right in the terminal: NOAA scales, Kp-index with a mini-graph, a three-day forecast, and chances to see the northern lights. Zero dependencies and no keys, npx solar-status. There is also a bot in Telegram: @solar_status_tgbot. About the connection between magnetic storms and well-being, it directly says that scientific data are contradictory.
→ github.com/vladzima/solar-status

tabs-tabs - a Chrome extension: sorts tabs by usage time, groups them by date, and unloads inactive ones from memory, but doesn't close them and doesn't touch groups you've assembled yourself.
→ github.com/vladzima/tabs-tabs

Swipehunt - Tinder for Product Hunt. Swipe launches: like, pass, save, and the feed adapts to what interests you. (The only non-OSS.)
→ swipe.computer


Why, Vladislav?

All these micro-projects first solved my own problems, and putting them in open source was inexpensive. If at least one is useful to you too - that's good. Stars on GitHub are accepted, bug reports even more so.



❗️❗️❗️❗️❗️❗️❗️❗️ / Not banned in the Russian Federation / Max