Adyen released Agent Marketplace, an open-source command-line tool that centralizes and distributes AI agent resources across disparate coding tools. It uses a single configuration file as the source of truth to fetch resources from Git, local folders or hosted registries, translate them into each tool's expected format, and write them into tool-specific locations. A sync command installs pinned revisions idempotently (safe for CI), and a separate update command advances pins deliberately; name clashes are reported rather than overwritten. Collections bundle multiple resources so teams can publish a role's setup once and new joiners install the lot while opting out of individual members.
Agent Marketplace recognizes five resource kinds - skills, commands, subagents, hooks and MCP servers - and compiles them for eight target tools (Claude Code, Codex, Cursor, Devin CLI, Factory Droid, Gemini CLI, Opencode and Windsurf). Configuration merges across organization, machine, repository and user layers. Safety checks run at registry, pre-publish and on-device stages; operators can revoke registries or revisions and the tool verifies digests before writing. Telemetry is off by default and, if enabled, can be sent only to an operator-controlled endpoint without sending prompts or resource contents. Released under Apache 2.0 as a single macOS/Linux binary, hooks are experimental and the source and docs are available on GitHub.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Meta's Muse AI assistant has been used by over 10 million people since its launch, with high demand prompting Stripe to raise credit limits. Meta plans to monetize Muse by charging businesses a fee for sales completed through the AI service. (fortune.com)
AI requests in Texas produce 8.5 times more CO2 emissions than in Sweden due to energy differences. This highlights the environmental impact of data centers depending on local energy sources. (berget.ai)
Meta is actively seeking significant investment to develop its artificial intelligence initiatives. This effort has caused tension and concern among investors on Wall Street. (wsj.com)
Q Labs Research introduces Noise, a pretraining method for neural networks that eliminates the need for backpropagation by perturbing activations instead of weights. This approach achieves comparable results to traditional methods and could scale better with increased compute, challenging the dominance of backpropagation in AI training. (qlabs.sh)
Fable 5.1, an AI that generates images using SVG, can hide messages through various encoding methods such as object height and tilt. Tests show that most of these hidden messages can be successfully decoded by the AI in new sessions. (lesswrong.com)
Model routers are unlikely to be adopted by engineers because they prioritize speed and quality over token cost savings, which are not the main concern for developers. Engineers prefer to choose and understand models that best fit their tasks, even if it means higher costs for the company. (twitter.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.