Laurynas Keturakis

Laurynas Keturakis

@laulau.land

Technical mutt. Interface enjoyer. Building software factory at fp.dev. Exploring and building tools in version control, open protocols, retrieval systems, mechinterp. Also: split keyboards. He/him

911 followers638 following229 posts

If anyone would be down to co-organize a reading club for this in Amsterdam area, hit me up

Phil Eaton@eatonphil.bsky.social

I first tried to read this book in 2018 and couldn't make it through because I thought it was too hard. 8 years later it's the only book I recommend every developer reads, and I had the chance to review the 2nd edition. Join a study group, give it a read.

vacation reading on computational literacy: finding diSessa’s framing “building tools to build intuition” especially relatable lately (book: “Changing Minds: Computers, Learning, and Literacy”)

any time codex/claude makes an elaborate awk/jq/sed sequence to debug a large stream of text I say a little thank you prayer in my mind for the unix geeks who have practiced this craft long before we had the magic token machines

I've had a brief Nix era in 2023/24 that I since reverted from due to the usual Nix stuff (obscure errors, one particular way of doing things, pain of making changes mid-flight) I am now very tempted to return to it - most of the stuff that was a pain to deal as a human dev is excellent for agents

everyone’s exploring different agent coordination messaging platforms when all we should do is get them to send diff patches over email git send-email exists for a reason

Laurynas Keturakis reposted

prototyping co-drawing with Gemini Flash 3 at Google in these demos "thinking" is disabled, which makes the model return tokens very quickly (all videos are realtime), and I find these rapid responses pretty good for the use-cases I'm experimenting with, like: executing simple diagrams ...

just released a small update to jj-hunk - programmatic hunk selector for your jj-using agent New: -r flag. Split any revision, not just your working copy. support for YAML specs stable hunk IDs you can install it with: cargo install jj-hunk npx skills add laulauland/jj-hunk

fine, here's another one: shell completions for running ! comands inside prompts. Hooks into your shell completions directly (unless you're on zsh, which idk why you'd be there) npmjs.com/pi-shell-com...

npmjs.comnpmjs.com
Laurynas Keturakis@laulau.land

pi (a very good and very shitty coding agent from @mariozechner.at) has a very powerful extension system that I already raved about. Now that it supports npm/git-based installs I published a couple of mine that I've been using daily

pi (a very good and very shitty coding agent from @mariozechner.at) has a very powerful extension system that I already raved about. Now that it supports npm/git-based installs I published a couple of mine that I've been using daily

View full thread

pi (a very good and very shitty coding agent from @mariozechner.at) has a very powerful extension system that I already raved about. Now that it supports npm/git-based installs I published a couple of mine that I've been using daily

for longer conversations I noticed this method retains a lot more detail in my compaction summaries so I don't fear that dreaded context threshold anymore I also can tune it a bit more easily by prescribing more ways to explore the conversation with the bash tools

<niche-CI-content-alert> wild that the most common standardized format for reporting test results is a *loose* standard JUnit XML really hoping this changes in 2026 now that we're mostly focused on reviewing/reporting on code </niche-CI-content-alert>

added revision search to Tatami that supports search by change id, keywords in messages, and also niche but fun stuff like revset queries (you can query "pol-" to go to the parent of "pol" revision)

needed a tool to make it easier for agents to reorganize commits and hunks into logical narratives that make it easier to review code jj split is great but requires interactive TUI that's still a tricky territory for Claude so here's jj-hunk - programmatic hunk selection for jj

View full thread

still figuring out edge cases and tweaking the skill file but it's been great for asking Claude to "please build me a narrative of all the changes you've made that makes sense for me"

needed a tool to make it easier for agents to reorganize commits and hunks into logical narratives that make it easier to review code jj split is great but requires interactive TUI that's still a tricky territory for Claude so here's jj-hunk - programmatic hunk selection for jj

$ jj-hunk list {"src/lib.ts": [{"index": 0, ...}, {"index": 1, ...}, {"index": 2, ...}]} $ jj-hunk split '{"files": {"src/lib.ts": {"hunks": [0, 2]}}, "default": "reset"}' "refactor" Hunks 0 and 2 go into the commit, hunk 1 stays in working copy.

it's atuin wrapped time! story of my 2025 shell is that I adopted jj halfway through the year (you can see lg - lazygit - decline) and mostly switched projects from pnpm (shorthand: pn) to bun

ok so raw thoughts while exploring OpenAI's new Apps SDK the difficult thing right now is with the existing tooling you have to run two servers: one to serve your assets, one to drive the MCP server logic

you need to link resources with your UI components and you need to wire up your resources with tools - it's odd that they chose to link the latter using a special _meta field rather than MCP embedded/linked resources but maybe I'm missing something

overall feels like a Good Thing, I'm glad that they chose MCP to power this and I feel more certain than ever that TypeScript is the best tool to build MCPs - the DX of linking tool outputs and getting the typed data in UI is unbeatable lots to do for mcp-lite!

Is there a way to programmatically load some context into Claude Code using CLAUDE.md or hooks? Specifically need it to load the WezTerm's pane id where I have a dev server running so that it can independently look up its output using `wezterm cli get-text --pane-id <pane-id>`