Notes to self · Marcus Michaels https://marcusmichaels.com/notes/ 2026-07-07T00:00:00Z Marcus Michaels Run two versions of the same npm package https://marcusmichaels.com/notes/run-two-versions-of-the-same-npm-package/ 2026-07-07T00:00:00Z How to run two versions of the same npm package at once. Install the new one under an npm alias, shim its binary under a name of its own, and point one script at it. The working example is TypeScript 7's RC. Running background agents, for humans https://marcusmichaels.com/notes/running-background-agents-for-humans/ 2026-07-05T00:00:00Z An AI agent that works in its own throwaway worktree, in the background, and comes back as a pull request you review. Ice cream flavoured Vim https://marcusmichaels.com/notes/ice-cream-flavoured-vim/ 2026-07-03T00:00:00Z Vim Scoops is a browser game that teaches Vim motions by steering an ice cream truck to its customers. The changelog that writes itself https://marcusmichaels.com/notes/local-llms-in-github-actions/ 2026-06-26T00:00:00Z Run a small language model on a GitHub Actions runner, with no API key and no data leaving the box. Our working example is a weekly changelog that writes itself and posts to Slack every Friday. Work on multiple branches at once with git worktrees https://marcusmichaels.com/notes/git-worktrees/ 2026-06-25T00:00:00Z A worktree is a separate working copy backed by the same repo. Keep your current checkout untouched whilst you build on another branch next door, then throw it away. Run everything in one terminal with screen https://marcusmichaels.com/notes/run-everything-in-one-terminal-with-screen/ 2026-06-18T00:00:00Z screen is a terminal multiplexer that holds every session in one window, keeps them alive when you walk away, and reattaches from anywhere. The big O cheatsheet https://marcusmichaels.com/notes/big-o-notation/cheatsheet/ 2026-06-17T00:00:00Z Every shape, data-structure costs, the recipe, and the survival table. The data structures to know by heart https://marcusmichaels.com/notes/big-o-notation/data-structures/ 2026-06-17T00:00:00Z What arrays, hash maps, trees, and heaps cost, and why you should "just use a Set". What Θ, Ω, and O really mean https://marcusmichaels.com/notes/big-o-notation/formal-definitions/ 2026-06-17T00:00:00Z The formal definitions; O, Ω, and Θ, and the one-line maths behind "drop the constants". What space complexity is https://marcusmichaels.com/notes/big-o-notation/space-complexity/ 2026-06-17T00:00:00Z The other axis; how much extra memory an algorithm needs, and the memory-for-speed trades you'll make. How I wish I was taught big O notation https://marcusmichaels.com/notes/big-o-notation/ 2026-06-16T00:00:00Z Big O, built in an ice cream van. A feel for how an algorithm's cost grows, then the notation behind it. How to work out the big O of your code https://marcusmichaels.com/notes/big-o-notation/reading-your-code/ 2026-06-16T00:00:00Z A three-step recipe for reading the big O off any function; find the loops, multiply or add, drop the constants. Why sorting is O(n log n) https://marcusmichaels.com/notes/big-o-notation/sorting-and-logarithms/ 2026-06-16T00:00:00Z Logarithms without fear, and why a good sort (and JS's .sort()) is O(n log n) in time. Which complexities are too slow https://marcusmichaels.com/notes/big-o-notation/too-slow/ 2026-06-16T00:00:00Z Where O(2ⁿ) and O(n!) come from, and the rule of thumb for what size of input each shape survives. What big O actually is https://marcusmichaels.com/notes/big-o-notation/what-big-o-is/ 2026-06-16T00:00:00Z n is the size of the input, and big O is how the work grows as it gets bigger. How big should a pull request be? https://marcusmichaels.com/notes/how-big-should-a-pull-request-be/ 2026-06-15T00:00:00Z Small, and about one thing. The size that keeps a change reviewable, and why that sets a team's pace. Undo almost anything with git reflog https://marcusmichaels.com/notes/git-reflog/ 2026-06-14T00:00:00Z A bad reset, lost commits, a branch you deleted by mistake. The reflog gets them back, because git rarely throws a commit away. Ask the history https://marcusmichaels.com/notes/grep-like-an-llm/ask-the-history/ 2026-06-12T00:00:00Z When the code won't explain itself, the commits usually will. The cheatsheet https://marcusmichaels.com/notes/grep-like-an-llm/cheatsheet/ 2026-06-12T00:00:00Z Every command from the series, in one place. Find the door https://marcusmichaels.com/notes/grep-like-an-llm/find-the-door/ 2026-06-12T00:00:00Z Entry points. Start from the text on screen, the names of files, and the lines around a match. Follow the data https://marcusmichaels.com/notes/grep-like-an-llm/follow-the-data/ 2026-06-12T00:00:00Z Read the types first, match the word rather than the substring, and find who writes a value. Grep like an LLM https://marcusmichaels.com/notes/grep-like-an-llm/ 2026-06-11T00:00:00Z How to find your way around any codebase with grep, git, and a method borrowed from an AI agent. Make a map, don't read https://marcusmichaels.com/notes/grep-like-an-llm/make-a-map/ 2026-06-11T00:00:00Z The heat map and the file list. Locating a feature before opening a single file. Playbook https://marcusmichaels.com/notes/grep-like-an-llm/playbook/ 2026-06-11T00:00:00Z Why maps beat reading, and the method in full before the modules begin. How this site is built https://marcusmichaels.com/notes/how-this-site-is-built/ 2026-06-10T00:00:00Z Markdown, a small build script, and modest-ui. The setup behind these notes.