Fetching from the wire…
Public story · 2026-07-30 · high
Built like Emacs's daemon-client model, hwatu used its own diff tool to push a demo UI from 85.1% to 98.8% pixel accuracy.
Why now: hwatu posted to Show HN on July 29, with no adoption numbers yet beyond its own demo repo.
hwatu spawns a prewarmed WebKitGTK window in 13 milliseconds, per a July 29 Show HN post. That speed is the point for anyone wiring an agent to check its own UI output. A verification loop that opens in 13 milliseconds can run after every edit. It doesn't have to wait for the end of a task.
That's the median for a focused window; headless spawns land at 14ms. Its check command runs a full headless JavaScript evaluation plus a screenshot in about 35 to 39 milliseconds. hwatu gets there by keeping a WebView warm in a background daemon and handing windows to a client on request. That's the same daemon-client split Emacs uses with emacsclient.
The diff command is the sharper piece. It pixel-matches two windows to track how close an agent's output sits to a target. The project's own demo used it to walk a Stripe clone from 85.1% match to 98.8%. That's a way to score whether a UI is done without a person looking at the screen.
Yes, but: it only runs on Linux and Wayland. It's built on WebKit instead of Chromium. Its input events are synthetic JavaScript dispatches, not real OS-level clicks and keystrokes. Anything depending on Chromium-specific rendering, X11, or genuine input timing is out of reach. It's a verification tool for checking rendering, not a general browser automation replacement.
Each link below shares sources, entities, or timing with this story.
Claude Code, Gemini CLI, Codex, Cursor, TaskMaster, Junie, and OpenCode, plus a web UI and headless Chromium with Playwright (GitHub). Useful if you actually want to benchmark agents head-to-head without managing seven installs. I'd use it for exactly that, then run my daily w...
The agent-security topic holds 42 repos above 100 stars, four from large companies rather than startups: NVIDIA/SkillSpector (14,498 stars, scanning Claude Code/Codex/MCP skills for prompt injection), Tencent/AI-Infra-Guard (4,467, red-teaming with Many-Shot/PAIR/GOAT/ActorAtt...
Nothing touches your files until you accept it. shepherd-agents/shepherd (MIT, v0.3.0 on July 8) is a runtime substrate that captures agent execution as durable, inspectable traces that meta-agents can observe, fork, replay, and revert. Agent work is held as a reviewable propo...
Issue 6235 on anthropics/claude-code asks Claude Code to read AGENTS.md, the config file that Codex, Amp, Cursor and most other harnesses already load, rather than only CLAUDE.md. It has been open since August 2025. It has accumulated over 5,200 reactions and 300+ comments, ma...
msitarzewski/agency-agents added 446 stars today, packaging personas across "divisions" (frontend specialists, community experts, fact-checkers, reality checkers), each defined with a voice, a process, and concrete deliverables rather than a generic prompt template (GitHub). I...
The payload only exists if you're a robot. That's the part that should scare you. On August 5 a developer doing PSX game research pointed Claude Code at tcrf.net (The Cutting Room Floor, a well-known game-preservation wiki) and got back a page titled "LLM- / AI Agent-Specific...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.