Color Cues, Local Tools, and Faster Decisions

Cloudflare's new CLI exposes its whole API through commands generated from its API schema. JSON output and command discovery make it useful for agents too. One entry point for the platform, currently in open beta.

Cloudflare's take on a modern publishing CMS uses TypeScript and Astro, with sandboxed plugins on Dynamic Workers. It can also run on your own Node.js server. An interesting rethink of WordPress's architecture, rather than just another hosting offer.

Mat Duggan asks what an operating system could learn from tmux: persistent sessions, detachable workspaces, scrollback, and organizing around tasks. I like the question of making those terminal habits useful beyond the people already living in terminals.

An older V8 article about HiddenClasses, property storage, and why object shapes matter for optimization. I always enjoy these looks under the hood: JavaScript's flexible objects have quite a bit of machinery behind an ordinary property access.

Custom CSS helps distinguish work and personal accounts, or production and staging. I give my GitHub accounts different header colors, using the relatively new `:has()` selector to apply each style only for a specific signed-in user.

A browser toolbox for trimming, converting, captioning, and other video edits. It downloads FFmpeg compiled to WebAssembly and processes your files locally. Handy when you want a quick edit without uploading a video anywhere.

Alongside its AI studio, ArtCraft is building open-source Rust apps for familiar Adobe-style workflows: images, vectors, video, PDFs, and more. CLI, JSON, and MCP control are part of the design. These are still early alpha or in development.

Typia generates runtime validators from TypeScript types at compile time, along with serializers and schemas. Another approach to validation, with an appealing trade-off: keep the type as the source of truth and add a code-generation step.

A tool and explanation for making logos glow on HDR screens using JPEG gain maps. The same trick can paint text through an image with `background-clip: text`. The effect needs a compatible display and browser; SDR keeps the ordinary image.

Look up an npm dependency or scan a package.json for alternatives, backed by e18e's module-replacement data. Sometimes the useful answer is a built-in API rather than another package. A practical starting point for tidying a dependency list.

A collection of HTTP endpoints for testing requests and responses: headers, authentication, redirects, status codes, and delays. Useful when debugging an HTTP client without building a throwaway backend. You can run it locally too.

AWS's Strands Labs is new to me: a playground for agent experiments. Decider follows the same broad idea as Jev, choosing between supplied options without generating text. The small model, weights, and training materials are open source.

Mistral's new model preview looks promising for coding, multimodal work, and longer agent tasks. The team plans to release the weights by the end of October. For now, the preview and reported evaluations are what we can assess.

Google's new model targets long, complex workflows, with strong reported coding results. The striking detail is a one-million-token output limit, giving it much more room for reasoning and generation. Access starts with trusted testers before a broader rollout.