🛠️ All DevTools

[Other] No-build, no-NPM, SSR-first JavaScript framework if you hate React, love HTML

[Other] I turned Markdown into a protocol for generative UI There&#x27;s a lot of work happening around both generative UI and code execution for AI agents. I kept wondering: how do you bring them together into a fully featured architecture? I built a prototype:<p>- Markdown as protocol — one stream carrying text, executable code, and data<p>- Streaming execution — code fences execute statement by statement as they stream in<p>- A mount() primitive — the agent creates React UIs with full data flow between client, server, and LLM<p>Let me know what you think!

[CLI Tool] Show HN: Agentic Copilot – Bring Claude Code, OpenCode, Gemini CLI into Obsidian Obsidian plugin that connects to CLI agents you already have installed. No built-in LLM integration, no API keys to configure in the plugin. It spawns your tool as a child process, pipes vault context into each prompt, and streams responses into a chat panel.<p>Supports Claude Code, Opencode, and any custom binary via a generic adapter. Adding a new agent is a single file. Free, proudly Open Source (MIT licensed).<p>Would love feedback on this for anyone that that tries it out.

[Other] A light-weight and powerful meta-prompting, context engineering and spec-driven development system for Claude Code by TÂCHES.

[Testing] Painless E2E Automation for Mobile and Web

[Other] Show HN: Ripl – A unified 2D/3D engine for Canvas, SVG, WebGPU, and the Terminal After several years, with a small hiatus in the middle, I&#x27;ve finally got Ripl to the point of being published. Ripl is a library for rendering 2D and 3D shapes to any context (canvas, SVG, WebGPU, and Terminal supported by default) using a single API. The library mimics the DOM as much as possible, replicating the event system, object graphing, CSS-like querying, gradients, and keyframe animations etc.<p>I also built a complete data visualization library using the core package which is available as @ripl&#x2F;charts. And yes, you can even render the charts to a terminal with about a 2-3 line code change :) (see the terminal demo)<p>Docs are available here: <a href="https:&#x2F;&#x2F;www.ripl.rocks" rel="nofollow">https:&#x2F;&#x2F;www.ripl.rocks</a> Demos are available here: <a href="https:&#x2F;&#x2F;www.ripl.rocks&#x2F;demos" rel="nofollow">https:&#x2F;&#x2F;www.ripl.rocks&#x2F;demos</a> Charts are avialable here: <a href="https:&#x2F;&#x2F;www.ripl.rocks&#x2F;docs&#x2F;charts" rel="nofollow">https:&#x2F;&#x2F;www.ripl.rocks&#x2F;docs&#x2F;charts</a><p>I&#x27;ve also built an interactive playground you can use to play around with it in realtime without having to install it from NPM etc. The playground is available here: <a href="https:&#x2F;&#x2F;www.ripl.rocks&#x2F;playground" rel="nofollow">https:&#x2F;&#x2F;www.ripl.rocks&#x2F;playground</a><p>The core library is quite stable and I&#x27;ll likely publish v1 in the coming weeks. The charts, 3D, and Terminal packages are still very experiemental.<p>I&#x27;d interested to hear what you all think of it.

[Testing] Bombadil: Property-based testing for web UIs by Antithesis

[Other] A sufficiently detailed spec is code

[CLI Tool] Cook: A simple CLI for orchestrating Claude Code

[Other] FastAPI-compatible Python framework with Zig HTTP core; 7x faster

[Other] Show HN: Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training I replicated David Ng&#x27;s RYS method (<a href="https:&#x2F;&#x2F;dnhkng.github.io&#x2F;posts&#x2F;rys&#x2F;" rel="nofollow">https:&#x2F;&#x2F;dnhkng.github.io&#x2F;posts&#x2F;rys&#x2F;</a>) on consumer AMD GPUs (RX 7900 XT + RX 6950 XT) and found something I didn&#x27;t expect.<p>Transformers appear to have discrete &quot;reasoning circuits&quot; — contiguous blocks of 3-4 layers that act as indivisible cognitive units. Duplicate the right block and the model runs its reasoning pipeline twice. No weights change. No training. The model just thinks longer.<p>The results on standard benchmarks (lm-evaluation-harness, n=50):<p>Devstral-24B, layers 12-14 duplicated once: - BBH Logical Deduction: 0.22 → 0.76 - GSM8K (strict): 0.48 → 0.64 - MBPP (code gen): 0.72 → 0.78 - Nothing degraded<p>Qwen2.5-Coder-32B, layers 7-9 duplicated once: - Reasoning probe: 76% → 94%<p>The weird part: different duplication patterns create different cognitive &quot;modes&quot; from the same weights. Double-pass boosts math. Triple-pass boosts emotional reasoning. Interleaved doubling (13,13,14,14,15,15,16) creates a pure math specialist. Same model, same VRAM, different routing.<p>The circuit boundaries are sharp — shift by one layer and the effect disappears or inverts. Smaller models (24B) have tighter circuits (3 layers) than larger ones (Ng found 7 layers in 72B).<p>Tools to find circuits in any GGUF model and apply arbitrary layer routing are in the repo. The whole thing — sweep, discovery, validation — took one evening.<p>Happy to answer questions.

[IDE/Editor] Show HN: Tmux-IDE, OSS agent-first terminal IDE Hey HN,<p>Small OSS project that i created for myself and want to share with the community. It&#x27;s a declarative, scriptable, terminal-based IDE focussed on agentic engineering.<p>That&#x27;s a lot of jargon, but essentially its a multi-agent IDE that you start in your terminal.<p>Why is that relevant? Thanks to tmux and SSH, it means that you have a really simple and efficient way to create your own always-on coding setup.<p>Boot into your IDE through ssh, give a prompt to claude and close off your machine. In tmux-ide claude will keep working.<p>The tool is intentionally really lightweight, because I think the power should come from the harnesses that you are working with.<p>I&#x27;m hoping to share this with the community and get feedback and suggestions to shape this project! I think that &quot;remote work&quot; is directionally correct, because we can now have extremely long-running coding tasks. But I also think we should be able to control and orchstrate that experience according to what we need.<p>The project is 100% open-source, and i hope to shape it together with others who like to work in this way too!<p>Github: <a href="https:&#x2F;&#x2F;github.com&#x2F;wavyrai&#x2F;tmux-ide" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;wavyrai&#x2F;tmux-ide</a> Docs: <a href="https:&#x2F;&#x2F;tmux.thijsverreck.com&#x2F;docs" rel="nofollow">https:&#x2F;&#x2F;tmux.thijsverreck.com&#x2F;docs</a>

[Code Quality] Show HN: I got tired of print(x.shape) so I built runtime type hints for Python As a beginner learning to build ML models, I found it annoying to have to keep printing tensor shapes every other line, having to step through the debugger to check where did I mess up the shapes again.<p>So I built Trickle, it takes the data that flows through your code, caches the types and display them inline (as if you have type annotations).<p>The idea is: &quot;Let types trickle from runtime into your IDE&quot;. You get types in Python without having the write them manually.<p>It works by rewriting your Python AST at import time — after every variable assignment, it inserts a lightweight call that records the type and value. No decorators, no code changes. Just run your script through trickle run python train.py and every variable gets its type visible.<p>One cool feature is Error snapshots, by toggling it in VSCode status bar, you can see the exact data that is in each variable when the error happened.<p>For AI agents, trickle-cli outputs the inline runtime types together with the source code so agent can better diagnose the issue.<p>For Jupyter notebooks: %load_ext trickle as your first cell, and every cell after is traced.<p>Quick try: pip install trickle-observe npm install -g trickle-cli code --install-extension yiheinchai.trickle-vscode<p><pre><code> trickle run python demo&#x2F;demo.py trickle hints demo&#x2F;demo.py </code></pre> Limitations: - Expect 2-5x slowdown — designed for development, not production.<p>Also supports JavaScript&#x2F;TypeScript (Express, Fastify, Koa, Hono) though the Python&#x2F;ML side is where I&#x27;ve focused most effort.<p>In the future, I imagine there to be potential for using this as runtime observability in production via probabilistic sampling of types. Now, we know the code, we know the data, which is all the information we need to debug.<p>Happy to answer questions

[Other] Google Engineers Launch "Sashiko" for Agentic AI Code Review of the Linux Kernel

[Other] An Open-Source Asynchronous Coding Agent

[Other] Parallel Perl – autoparallelizing interpreter with JIT

[Other] Show HN: N0x – LLM inference, agents, RAG, Python exec in browser, no back end Built this because I was tired of every AI tool shipping my data to someone else server n0x runs the full stack LLM inference via WebGPU, autonomous ReAct agents, RAG over your own docs, sandboxed Python execution via Pyodide all inside a single browser tab. No account No keys No backend Models download once, cache in IndexedDB permanently. Biggest challenge was context window budgeting for the agent loop and making the WASM vector search non-blocking. Happy to talk architecture.<p>GitHub: <a href="https:&#x2F;&#x2F;github.com&#x2F;ixchio&#x2F;n0x" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;ixchio&#x2F;n0x</a> | Live demo: <a href="https:&#x2F;&#x2F;n0x-three.vercel.app" rel="nofollow">https:&#x2F;&#x2F;n0x-three.vercel.app</a>

[Other] Launch an autonomous AI agent with sandboxed execution in 2 lines of code

[CLI Tool] Show HN: Horizon – GPU-accelerated infinite-canvas terminal in Rust Tabs, splits, and tmux work fine until you have several projects open with logs, tests, and long-running shells. I kept rebuilding context instead of resuming work. Horizon puts shells on an infinite canvas. You can arrange them into workspaces and reopen later with layout, scrollback, and history intact.<p>Built in 3 days with Claude&#x2F;Codex, dogfooding the workflow as I went. Feedback and contributions welcome.

[Other] Edge.js: Run Node apps inside a WebAssembly sandbox