🛠️ Hacker News Tools

[CLI Tool] Show HN: Amla Sandbox – WASM bash shell sandbox for AI agents WASM sandbox for running LLM-generated code safely.<p>Agents get a bash-like shell and can only call tools you provide, with constraints you define. No Docker, no subprocess, no SaaS — just pip install amla-sandbox

[Other] Show HN: Cicada – A scripting language that integrates with C I wrote a lightweight scripting language that runs together with C. Specifically, it&#x27;s a C library, you run it through a C function call, and it can callback your own C functions. Compiles to ~250 kB. No dependencies beyond the C standard library.<p>Key language features: * Uses aliases not pointers, so it&#x27;s memory-safe * Arrays are N-dimensional and resizable * Runs scripts or its own &#x27;shell&#x27; * Error trapping * Methods, inheritance, etc. * Customizable syntax

[Other] Tomo: A statically typed, imperative language that cross-compiles to C [video]

[DevOps] Show HN: Autonomous recovery for distributed training jobs Hi HN! We’re TensorPool. We help companies access and optimize large scale compute for training foundation models.<p>The Problem<p>It’s been almost a year since we’ve finished YC, and we’ve just crossed 100,000 multinode training GPU hours run on our platform.<p>On those training runs, we’ve seen countless 3am job crashes because of issues like an Xid error from a flaky GPU or an S3 timeout that corrupted a checkpoint save. By the time you wake up and notice, you&#x27;ve lost 8+ hours of compute. You scramble to diagnose the issue, manually restart from the last checkpoint, and hope it doesn&#x27;t happen again. Rinse and repeat.<p>For training runs that take days to weeks, this constant babysitting is exhausting and expensive. The research iteration cycles lost can also make or break a model release (especially for short reservations).<p>What We Built<p>This agent monitors your training jobs and autonomously recovers them when things go wrong. It works with Kubernetes, Slurm, and TensorPool Jobs.<p>We originally built the TensorPool Agent as an internal tool to help us debug failures with our own customers. Over time, we realized its performance was so good that we could automate the entire triage process. We&#x27;re now releasing a public beta for people to use.<p>Best case: The TensorPool Agent detects the failure, diagnoses the root cause, fixes it, and restarts your job from the last checkpoint – all while you sleep ;)<p>Worst case: If the TensorPool agent can&#x27;t fix the issue automatically, it delivers a preliminary RCA and a list of actions it attempted, giving you a head start on debugging.<p>How It Works<p>1) Registration – You provide credentials to your job scheduler via our dashboard. Perms are granted on a whitelist basis; you explicitly control what actions the agent can take.<p>2) Monitoring – The agent continuously monitors your job for failure conditions.<p>3) Recovery – On failure, the agent analyzes logs and attempts to diagnose the issue. If successful, it restarts the job from the last checkpoint and resumes monitoring. If not, you get an alert with full context.<p>Target Failure Modes<p>The agent is specifically designed for runtime errors that occur deep into training, like:<p>- CUDA OOM: Memory leaks, gradient explosions<p>- Xid errors: GPU hardware faults (Xid 79, 63, 48, etc.)<p>- Distributed communication failures: NCCL timeouts, rank failures<p>- Storage I&#x2F;O errors: Checkpoint corruption<p>- Network issues: S3 request timeouts on mounted object storage

[Other] Show HN: vind – A Better Kind (Kubernetes in Docker)

[Other] Show HN: I made a dual-bootable NixBSD (NixOS and FreeBSD) image I&#x27;ve been working on getting NixBSD (Nix package manager + FreeBSD) to boot alongside NixOS on a shared ZFS pool. The result is a &lt;2GB disk image you can try in QEMU or virt-manager.<p>What works:<p><pre><code> - GRUB chainloads FreeBSD&#x27;s bootloader - Both systems share a ZFS pool - Everything is defined in a single Nix flake - Fully reproducible builds (some dependencies are now cached on Cachix) </code></pre> Planned:<p><pre><code> - Support native compilation of NixBSD (currently cross-compiled on Linux) - Many shortcuts were taken to get this working, needs lots of cleanup - Add a semi-automated installer like nixos-wizard </code></pre> Try it:<p><pre><code> qemu-system-x86_64 -enable-kvm -m 2048 \ -bios &#x2F;usr&#x2F;share&#x2F;ovmf&#x2F;OVMF.fd \ -drive file=nixos.root.img,format=raw </code></pre> Login: nixos&#x2F;nixos or root&#x2F;toor<p>The hardest parts were getting mounts working at boot, making the bootloader setup idempotent, and debugging early init. This disk image could potentially work on a USB stick with a bit more work.<p>This is very much experimental. My goal is to eventually produce a proper NixBSD installation ISO and consolidate all configuration into one repository while still consuming upstream NixBSD as a flake.<p>Download: <a href="https:&#x2F;&#x2F;github.com&#x2F;jonhermansen&#x2F;nixbsd-demo&#x2F;releases&#x2F;tag&#x2F;build-1" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;jonhermansen&#x2F;nixbsd-demo&#x2F;releases&#x2F;tag&#x2F;bui...</a><p>Feel free to leave feedback here or on GitHub! Thanks!

[Other] Claude Code daily benchmarks for degradation tracking

[Other] Render Mermaid diagrams as SVGs or ASCII art

[Other] Mermaid ASCII: Render Mermaid diagrams in your terminal <a href="https:&#x2F;&#x2F;github.com&#x2F;AlexanderGrooff&#x2F;mermaid-ascii" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;AlexanderGrooff&#x2F;mermaid-ascii</a>

[Other] pg_tracing: Distributed Tracing for PostgreSQL

[Other] Show HN: Cursor for Userscripts I’ve been experimenting with embedding an Claude Code&#x2F;Cursor-style coding agent directly into the browser.<p>At a high level, the agent generates and maintains userscripts and CSS that are re-applied on page load. Rather than just editing DOM via JS in console the agent is treating the page, and the DOM as a file.<p>The models are often trained in RL sandboxes with full access to the filesystem and bash, so they are really good at using it. So to make the agent behave well, I&#x27;ve simulated this environment.<p>The whole state of a page and scripts is implemented as a virtual filesystem hacked on top of browser.local storage. URL is mapped to directories, and the agent starts inside this directory. It has the tools to read&#x2F;edit files, grep around and a fake bash command that is just used for running scripts and executing JS code.<p>I&#x27;ve tested only with Opus 4.5 so far, and it works pretty reliably. The state of the file system can be synced to the real filesystem, although because Firefox doesn&#x27;t support Filesystem API, you need to manually import the fs contents first.<p>This agent is <i>really</i> useful for extracting things to CSV, but it&#x27;s also can be used for fun.<p>Demo: <a href="https:&#x2F;&#x2F;x.com&#x2F;ichebykin&#x2F;status&#x2F;2015686974439608607" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;ichebykin&#x2F;status&#x2F;2015686974439608607</a>

[Other] Show HN: A MitM proxy to see what your LLM tools are sending I built this out of curiosity about what Claude Code was actually sending to the API. Turns out, watching your tokens tick up in real-time is oddly satisfying.<p>Sherlock sits between your LLM tools and the API, showing you every request with a live dashboard, and auto-saved copies of every prompt as markdown and json.

[Other] Mousefood – Build embedded terminal UIs for microcontrollers

[Other] Oban, the job processing framework from Elixir, has come to Python

[Other] Show HN: Config manager for Claude Code (and others) – rules, MCPs, permissions I use Claude Code across multiple projects with different conventions and some shared repos just as it so happens to be the real world. Managing the config files (.claude&#x2F;rules&#x2F;, mcps.json, settings.json) by hand got tedious, so I built a local web UI for it.<p>This one started out as claude-config but migrated to coder-config as I&#x27;m adding others (Gemini, AG, Codex, etc).<p>Main features: - Visual editor for rules, permissions, and MCP servers - Project registry to switch between codebases - &quot;Workstreams&quot; to group related repos (frontend + API + shared libs) with shared context - Auto-load workstreams on cd to included folders - Also supports Gemini CLI and Codex CLI<p>Install: npm install -g coder-config coder-config ui # UI at http:&#x2F;&#x2F;localhost:3333 coder-config ui install # optionally, autostart on MacOS<p>It can also be installed as a PWA and live in your taskbar.<p>Open source, runs locally, no account needed. Feedback and contributions welcome!<p>Sorry, haven&#x27;t had any chance to test on other OSes (linux&#x2F;windows)

[API/SDK] Show HN: Sandbox Agent SDK – unified API for automating coding agents We’ve been working with automating coding agents in sandboxes as of late. It’s bewildering how poorly standardized and difficult to use each agent varies between each other.<p>We open-sourced the Sandbox Agent SDK based on tools we built internally to solve 3 problems:<p>1. Universal agent API: interact with any coding agent using the same API<p>2. Running agents inside the sandbox: Agent Sandbox provides a Rust binary that serves the universal agent API over HTTP, instead of having to futz with undocumented interfaces<p>3. Universal session schema: persisting sessions is always problematic, since we don’t want the source of truth for the conversation to live inside the container in a schema we don’t control<p>Agent Sandbox SDK has:<p>- Any coding agent: Universal API to interact with all agents with full feature coverage<p>- Server or SDK mode: Run as an HTTP server or with the TypeScript SDK<p>- Universal session schema: Universal schema to store agent transcripts<p>- Supports your sandbox provider: Daytona, E2B, Vercel Sandboxes, and more<p>- Lightweight, portable Rust binary: Install anywhere with 1 curl command<p>- OpenAPI spec: Well documented and easy to integrate<p>We will be adding much more in the coming weeks – would love to hear any feedback or questions.

[Other] Show HN: I built a small browser engine from scratch in C++ Hi HN! Korean high school senior here, about to start CS in college.<p>I built a browser engine from scratch in C++ to understand how browsers work. First time using C++, 8 weeks of development, lots of debugging—but it works!<p>Features:<p>- HTML parsing with error correction<p>- CSS cascade and inheritance<p>- Block&#x2F;inline layout engine<p>- Async image loading + caching<p>- Link navigation + history<p>Hardest parts:<p>- String parsing(html, css)<p>- Rendering<p>- Image Caching &amp; Layout Reflowing<p>What I learned (beyond code):<p>- Systematic debugging is crucial<p>- Ship with known bugs rather than chase perfection<p>- The Power of &quot;Why?&quot;<p>~3,000 lines of C++17&#x2F;Qt6. Would love feedback on code architecture and C++ best practices!<p>GitHub: <a href="https:&#x2F;&#x2F;github.com&#x2F;beginner-jhj&#x2F;mini_browser" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;beginner-jhj&#x2F;mini_browser</a>

[Other] Show HN: An extensible pub/sub messaging server for edge applications hi there! i’ve been working on a project called Narwhal, and I wanted to share it with the community to get some valuable feedback.<p>what is it? Narwhal is a lightweight Pub&#x2F;Sub server and protocol designed specifically for edge applications. while there are great tools out there like NATS or MQTT, i wanted to build something that prioritizes customization and extensibility. my goal was to create a system where developers can easily adapt the routing logic or message handling pipeline to fit specific edge use cases, without fighting the server&#x27;s defaults.<p>why Rust? i chose Rust because i needed a low memory footprint to run efficiently on edge devices (like Raspberry Pis or small gateways), and also because I have a personal vendetta against Garbage Collection pauses. :)<p>current status: it is currently in Alpha. it works for basic pub&#x2F;sub patterns, but I’d like to start working on persistence support soon (so messages survive restarts or network partitions).<p>i’d love for you to take a look at the code! i’m particularly interested in all kind of feedback regarding any improvements i may have overlooked.

[CLI Tool] Show HN: PNANA - A TUI Text Editor I’d like to share PNANA , a lightweight TUI editor built with C++ and FTXUI that I’ve been building for personal use and now open-sourced. It’s a minimal, fast terminal-based editor focused on simple coding and editing workflows—no bloated features, just the core functionality for terminal-centric use cases.<p><a href="https:&#x2F;&#x2F;github.com&#x2F;Cyxuan0311&#x2F;PNANA" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;Cyxuan0311&#x2F;PNANA</a><p>Key pragmatic features<p>Lightweight C++ core with FTXUI for smooth TUI rendering, fast startup and low resource usage<p>Basic but solid editing capabilities (syntax highlighting, line numbering, basic navigation)<p>Simple build process with minimal dependencies, easy to compile and run on Linux&#x2F;macOS terminals<p>Early LSP integration support for basic code completion (still polishing, but functional for common languages)<p>It’s very much an early-stage project—I built it to scratch my own itch for a minimal, self-built TUI editor and learn C++&#x2F;FTXUI along the way. There are definitely rough edges (e.g., some LSP kinks, limited customization), and it’s not meant to replace mature editors like Vim&#x2F;Nano—just a small open-source project for folks who like minimal terminal tools or want to learn TUI development with C++.<p>Any feedback, bug reports, or tiny suggestions are super welcome. I’m slowly iterating on it and would love to learn from the HN community’s insights. Thanks for taking a look!

[Other] Show HN: SHDL – A minimal hardware description language built from logic gates Hi, everyone!<p>I built SHDL (Simple Hardware Description Language) as an experiment in stripping hardware description down to its absolute fundamentals.<p>In SHDL, there are no arithmetic operators, no implicit bit widths, and no high-level constructs. You build everything explicitly from logic gates and wires, and then compose larger components hierarchically. The goal is not synthesis or performance, but understanding: what digital systems actually look like when abstractions are removed.<p>SHDL is accompanied by PySHDL, a Python interface that lets you load circuits, poke inputs, step the simulation, and observe outputs. Under the hood, SHDL compiles circuits to C for fast execution, but the language itself remains intentionally small and transparent.<p>This is not meant to replace Verilog or VHDL. It’s aimed at: - learning digital logic from first principles - experimenting with HDL and language design - teaching or visualizing how complex hardware emerges from simple gates.<p>I would especially appreciate feedback on: - the language design choices - what feels unnecessarily restrictive vs. educationally valuable - whether this kind of “anti-abstraction” HDL is useful to you.<p>Repo: <a href="https:&#x2F;&#x2F;github.com&#x2F;rafa-rrayes&#x2F;SHDL" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;rafa-rrayes&#x2F;SHDL</a><p>Python package: PySHDL on PyPI<p>To make this concrete, here are a few small working examples written in SHDL:<p>1. Full Adder<p>component FullAdder(A, B, Cin) -&gt; (Sum, Cout) {<p><pre><code> x1: XOR; a1: AND; x2: XOR; a2: AND; o1: OR; connect { A -&gt; x1.A; B -&gt; x1.B; A -&gt; a1.A; B -&gt; a1.B; x1.O -&gt; x2.A; Cin -&gt; x2.B; x1.O -&gt; a2.A; Cin -&gt; a2.B; a1.O -&gt; o1.A; a2.O -&gt; o1.B; x2.O -&gt; Sum; o1.O -&gt; Cout; }</code></pre> }<p>2. 16 bit register<p># clk must be high for two cycles to store a value<p>component Register16(In[16], clk) -&gt; (Out[16]) {<p><pre><code> &gt;i[16]{ a1{i}: AND; a2{i}: AND; not1{i}: NOT; nor1{i}: NOR; nor2{i}: NOR; } connect { &gt;i[16]{ # Capture on clk In[{i}] -&gt; a1{i}.A; In[{i}] -&gt; not1{i}.A; not1{i}.O -&gt; a2{i}.A; clk -&gt; a1{i}.B; clk -&gt; a2{i}.B; a1{i}.O -&gt; nor1{i}.A; a2{i}.O -&gt; nor2{i}.A; nor1{i}.O -&gt; nor2{i}.B; nor2{i}.O -&gt; nor1{i}.B; nor2{i}.O -&gt; Out[{i}]; } }</code></pre> }<p>3. 16-bit Ripple-Carry Adder<p>use fullAdder::{FullAdder};<p>component Adder16(A[16], B[16], Cin) -&gt; (Sum[16], Cout) {<p><pre><code> &gt;i[16]{ fa{i}: FullAdder; } connect { A[1] -&gt; fa1.A; B[1] -&gt; fa1.B; Cin -&gt; fa1.Cin; fa1.Sum -&gt; Sum[1]; &gt;i[2,16]{ A[{i}] -&gt; fa{i}.A; B[{i}] -&gt; fa{i}.B; fa{i-1}.Cout -&gt; fa{i}.Cin; fa{i}.Sum -&gt; Sum[{i}]; } fa16.Cout -&gt; Cout; } }</code></pre>