Show HN for April 30, 2026

I originally was just messing with pi-autoresearch. Gave it a sample task to build the most portable coding agent.

First cut was 6 KB of shell. Great for one-shots, unusable interactively. I was shocked it actually worked.

Started building up -- adding features — but with a self-imposed rule: no new dependencies, and sub 500 LOC. This thing had to be truly portable. Just sh, curl, awk. System primitives only.

Which means I did some genuinely disgusting things in awk, including JSON parsing and the OpenAI Responses tool loop with reasoning items carried across turns.

It's now ~400 lines. In the box: Anthropic + OpenAI, 7 tools (bash, read, write, edit, grep, find, ls), REPL, auto-compaction, checkpoint/resume, pipe mode, 90 no-API tests. Not in the box: TUI, streaming, images, OAuth, Windows, dignity.

Two honest things:

1. I stole/modified the system prompt and the architecture. Pi/Claude/Codex wrote the awk. I cannot read most of this code. This wasn't possible for me a year ago.

2. Heavily inspired by Pi (pi.dev) — same 7-tool surface, same exact-text edit model. Credit where it's due. Pi is awesome -- you should probably use them.

The agent loop itself is tiny. Almost everything else in a "real" agent CLI is DX and hardening. You can probably build your own harness exactly how you like it. Mario Zechner's AI Engineer talk on taking back control of your tools nudged me here.

The name is because it's a .sh file. The other thing it sounds like is, regrettably, also accurate.

In light of recent news about an agent deleting a production database, I thought now would be a good time to share this.

As the use of AI tools in production is becoming more common, sadly so will the high profile incidents like the one mentioned.

Fewshell is a terminal agent specifically designed to avoid this.

There is no setting to enable command auto-approval. This is by-design, so that the user never has to second-guess or worry about accidentally having it enabled.

Originally my intention was to build an AI mobile terminal to make typing shell commands easy. But with so many mobile-enabled 'claw' agents being available, I decided to make Fewshell the opposite of an autonomous agent.

Please star if you like, let me know what you think. Happy to answer questions.

About me: I'm an ex Amazon Sr. SDE for Alexa AI, and currently am working in AI safety research for agentic RLVR. I use this tool to run and check on my lab experiments.

Be horse.

AgentRQ is a (optionally) human-in-the-loop, self learning closed loop task manager for agents. Agents can create and schedule tasks for themself and work on them on their own schedule.

In high level it comes with one supervisor MCP that controls workspaces(worker agents) and unlimited number of isolated workspace MCPs (self learning agents).

Each workspace/agent has a mission/persona for the agent. And self-learning-loop note.

I am using it about 6 weeks in production, and completed more than 500 tasks. I just released the opensource version(as is in production) under Apache 2.0 license.

Currently it supports Gemini CLI and Claude code. I am going to extend support all major agents soon.

Happy to answer any questions.

Hi HN! Flocklist.app is a minimalist, private dependency tracker.

I mostly just built this for myself, since I would sit down to work, look at my to-do list, and I couldn't do 2/3 of the tasks. None of the solutions to this really cut it, either because they are a little clunky or are enterprise scale (like Jira, Linear, you know the rest). I figure I can't be the only one with this problem, so I built this free app to fix it. (It also doesn’t store tasks in the cloud.)

I would love to hear any feedback you have!

Fuzzy opcode windows. You don't need an exact number of vibes, just roughly right.

Hi HN,

I built ukbinday to provide a nicer way to find your bin collection days and actually remember when to put them out.

It shows your upcoming collections and generates a calendar that updates automatically as the schedule comes out. Coverage is at about 85% of the UK so far and I'm working on getting that to 100%. There is also an API for the Home Automation community.

This was built on top of two amazing open-source projects without which this would not exist: https://github.com/robbrad/UKBinCollectionData & https://github.com/mampfes/hacs_waste_collection_schedule Thank you to all the contributors of those repositories!

Two years ago I sat down to build a chess engine for my kids' Christmas present. Wanted it to play badly enough to be beatable but not randomly bad: different styles so they could choose what to face.

While building those styles I noticed the choices started looking like actual historical players' tendencies. Went looking. Turns out: given enough games from one person, you can extract enough of their decision pattern to reproduce it.

Two years later, 41 players, ~3,000 games each. Tal sacrifices unsoundly. Morphy attacks like it's 1850. Capablanca grinds endgames. Same engine underneath, visibly different choices each time.

playchessgate.com: browser, no signup. The kids never got their Christmas present.

A multiplayer hommage to Jean-Michel Sanejouand's Jeu de Topo (1963), a game with no winner or loser where partners arrange stones until both points of view are satisfied.

Here, stones become colored points in a game based on Voronoi diagrams. The original opposing points of view become each player's screen: everyone shares the same points, but phones, tablets, and browser windows frame the space differently, so each player is likely seeing a different Voronoi view.

I first posted about DD Photos 50 days ago[1], when I first open-sourced it. A lot of improvements have been made since then (hero image, site/album passwords, ngnix support, UI refinements, custom HTML/CSS, s3 deployments, and more).

However, I feel two significant additions are worth sharing (Docker and Surge).

Using DD Photos directly from source is very developer-heavy. You have to clone the repo and install Go, Node, libvips and all the dependencies. It's a lot to get started.

I thought why not pre-package everything all set to go? So DD Photos now has "Docker mode". An init command does the scaffolding, installing a 'ddphotos' wrapper script, and then there are a handful of simple commands to build the site.

mkdir ~/my-ddphotos

docker run --rm -v ~/my-ddphotos:/ddphotos dougdonohoe/ddphotos init

cd ~/my-ddphotos

./ddphotos photogen # resize images and create index files

./ddphotos run # run dev server at http://localhost:5173

./ddphotos build # build static site

./ddphotos serve # serve static site via Apache at http://localhost:8000

The second barrier is deploying a site. I use AWS s3/CloudFront/Route53 - which again is non-trivial to setup. I then discovered surge.sh, a free and easy way to deploy static sites in minutes. So I added support for it:

./ddphotos export --copy

surge --domain my-unique-site.surge.sh export/my-photos

Result of above: https://ddphotos-test-docker.surge.sh/

Anyway, the simplicity of this feels like a game-changer, which is why I wanted to share.

Code: https://github.com/dougdonohoe/ddphotos

[1] https://news.ycombinator.com/item?id=47322838

This is a C port of Theori/Xint's Copy Fail PoC (https://copy.fail/, 2026-04-29). The original depended on Python and embedded x86_64 shellcode. I managed to reimplement the shellcode as portable C using nolibc borrowed from the kernel, and now the whole thing can be compiled for & works on any arch supported by nolibc.

Hi HN, I’m Hal, the CTO at App Dev for All. I wanted to share a technical problem we worked on over the past year and how we approached it.

We’ve been building Code on the Go, a full-featured IDE that runs entirely on an Android phone. No laptop, no ADB connection, no cloud build server. It compiles projects locally on the device using Gradle, supports Java and Kotlin with LSP, and includes a debugger that runs on the same phone as the app being tested.

The most interesting and challenging part ended up being the debugger. The Android OS has a rigorous security model, which can get in the way of traditional inter-process communication. Android debugging assumes ADB, which assumes two machines. We bypassed ADB entirely, attaching the JDWP agent to the target process at launch and routing its output to our debugger over a local socket. We used a scoped adaptation of the Shizuku project to get the necessary system access without requiring root.

We also had a few other technical challenges with Code on the Go: Sketch-to-UI (generates Android XML from a photo of a hand-drawn layout, runs fully offline with Yolo), an optional Gemini-powered coding agent (opt-in, requires your own API key), and a plugin system with isolated class loaders.

One of our pre-release community members has used it to build and publish a Sinhala/English keyboard app to the Play Store, built entirely on his phone. This served as our test case for Play Store compatibility.

We are a philanthropic venture. No ads, no tracking, no subscription. License is GPLv3.

APK: http://appdevforall.org/codeonthego Source: http://github.com/appdevforall/CodeOnTheGo

Happy to answer questions on the implementation.

If you are a writer or someone who likes to share news, this could help make sharing articles to your social media following much easier.

Just highlight a quote, right click → Create Image and it generates a clean vertical card with the quote, the publication, the author, and the article title. The image can be either a 1x1 square or 9x16 vertical which is perfect for social media stories.

We're open sourcing Nimbalyst, a multi-agent, visual workspace for building with Claude Code, Codex, and Opencode (alpha). It pairs parallel session management with WYSIWYG editors so you and your agents can work on the same files at the same time.

The core idea: agents edit files, you edit files, and Nimbalyst makes that collaboration legible. Built-in WYSIWYG editors cover markdown, mockups, Excalidraw diagrams, data models, spreadsheets, and code (Monaco). Every edit is tracked back to the session that made it, with red/green WYSIWYG markdown diffs for review and approval. A session Kanban lets you search, resume, and manage parallel runs, and link sessions to files (and back). A lightweight tracker (alpha) keeps plans, bugs, tasks, and custom items in the same workspace the agent is working in — agents create, update, and execute them, and you see the same view they do. The extension system makes any file type pluggable; current extensions include Excalidraw, mind maps, CSV spreadsheets, a data model designer, mockups, PDF viewing, and a SQLite browser, and they're exposed to the agents too. Worktrees, workstreams, visual git management, and agent-driven commit proposals round out the developer side, and an iOS app handles remote session management (Android in development). Currently supports Claude Code, Codex, Opencode (alpha), and Copilot (alpha), with more planned.

It's local-first: files stay on your filesystem in open formats (markdown, excalidraw, html). The local app is MIT licensed and free; team collaboration features are AGPL and will be paid. Tech stack: Electron, React, Jotai, PGLite (Postgres in WASM), Lexical, Monaco, Playwright; collaboration runs on Cloudflare Workers with Durable Objects.

https://github.com/Nimbalyst/nimbalyst

https://nimbalyst.com/

We'd love your feedback and contributions.

So this all started two weeks ago as a stupid inside joke with friends in a group chat where we got a chatgpt powered bot to write plugins for itself and told it to build one that would make it fetch a random image from Wikipedia and then place words from the same article onto it. Often times it would just return mangled garbage or be completely incoherent but every now and then something astoundingly funny or profound would break through (if you have that certain 2005 idiotic internet humor that we/I certainly suffer from).

And well a sleepless weekend and two maxed out copilot subscriptions later, what I can only describe as antipattern for the pure sake of having dumb fun actually went live. For an added bonus, some banger keygen chiptunes are highly reccomended for browsing. Might put that in at some point in the future...

And on the incredibly minuscule chance that you even know or remember ][CdK][… sup bucko, we still ain’t dead.

I build a small tool for Mac to keep it work when the lid closed. It allows you to close the lid without worrying interupt your agents.

This began last year as a Python tool I wrote for my own use. I hate the "weighted shuffle" that Spotify uses. It can't handle experimental genres or music with generally low play counts. Also, when playing playlists with full albums, it rarely chooses any deep cuts. I wanted something with truly random shuffling from playlists of complete albums of niche music.

In building that, I found that I could make really interesting mixes of cross-genre music, especially from playlists of friends.

The built-in Spotify mixing forces you into a funnel of connecting Spotify membership collaboration and is terrible for using the playlists of interesting strangers. It also wants to use AI on everything and also modifies the original playlists.

I spent the last couple of months building out a web app for it. I wanted to explore three.js and Svelte more. It was great for that. Although I used Claude fairly often, I backed off because I actually wanted to learn the libraries I was using.

It's free. I hope the Spotify API rate-limiting holds up to general use like this.

I showed an earlier version of this here a bit over two years ago. Since then, Reli has grown a lot.

Reli is a sampling profiler and VM state inspector for PHP, written in PHP, that inspects running PHP processes from the outside.

In 0.12.0, tracing, memory analysis, monitoring, and runtime inspection all expanded quite a bit.

I would be especially interested in feedback on whether the memory-analysis, watch, and variable-inspection direction looks useful.

I've been building this solo for a while now. I usually prefer writing code, but Unity's shader graph pulled me in, I ended up building all the visuals with custom shaders, which gave the game its look.

The concept is a reverse city-builder: instead of expanding and consuming, you restore a wasteland back into a thriving ecosystem using robots.

Would love any feedback, both on the technical side and the game itself. Free demo just went live on Steam yesterday.