Notes

@code_kartik rebuilt BetterShot as a native Swift 6 + SwiftUI macOS screenshot tool aimed at anyone who wants a CleanShot X alternative without another subscription. It is local-first, skips web views and Tauri, and starts with the core capture flow: region, fullscreen, and window screenshots for free.

@insik0_0han built han-monorepo-template, a Better-T-Stack-derived TypeScript monorepo that pairs React web, Expo native, and Hono on Cloudflare Workers behind a shared oRPC contract. It standardizes the toolchain around Vite+, adds i18n and architecture linting, and includes a gated self-evolving AI workflow where the coding agent proposes improvements to its own instructions and skills through PRs.

@zeke points to macOS App Skills, Fayaz Ara's growing skill set for AI coding agents building native Mac apps with SwiftUI and AppKit. It packages the underdocumented parts of macOS work into reusable playbooks, from command-line Xcode builds to Sparkle auto-update wiring and Tahoe-era settings UI patterns, making it a practical bridge from web-agent workflows into real desktop software.

@kathyyliao says Cloudflare Workers can now call Browser Run Quick Actions directly through env.BROWSER.quickAction(), exposing HTML rendering, screenshots, Markdown extraction, PDFs, and related browser outputs as a native binding instead of a token-based API. Small surface area, but a useful step toward making browser automation feel like standard Workers infrastructure.

@ParthJadhav8 points to react-doctor, a simple agent-ready React maintenance loop: run the CLI, let your coding agent fix issues, and keep iterating until the app scores 100. Lightweight framing, but the underlying idea is useful: package codebase health checks into a single target an agent can repeatedly evaluate and improve.

@mr_r0b0t points to Webwright, a Microsoft terminal-native browser agent framework built on Playwright. The notable design choice is reuse: agents generate scripts, rerun them in fresh sessions, preserve logs and screenshots in the workspace, and turn one-off browser tasks into repeatable programs.

@__morse introduced Holocron, an open source Mintlify-compatible docs generator built as a Vite plugin. It uses MDX and docs.json, supports search, OpenAPI docs, and AI-readable exports, and can be deployed on Vercel, Cloudflare Workers, Docker, or a self-hosted setup.

@yoginth shipped bigRAG updates that make the self-hosted RAG platform more practical for real document pipelines: scheduled S3 and Cloudflare R2 sync plus realtime ingestion progress over sockets. The project already handles ingestion, vector search, and Turbopuffer-backed retrieval; this release tightens the loop for teams feeding docs from object storage into production knowledge bases.

@harrysolovay points to effect-cf, Daniel van der Merwe's Effect-native toolkit for Cloudflare Workers. It wraps Workers, Durable Objects, KV, bindings, and Durable Object storage in Effect's context, layer, and effect abstractions, keeping Cloudflare-specific runtime concerns at the edges instead of leaking through application code.

@xogotapp is bringing Xogot for Mac, its Apple-native take on Godot, into open beta. The pitch is a more Mac-like game development workflow with Xcode-inspired navigation, built-in Apple deployment, simulator support, and less friction moving projects onto iPhone, iPad, and macOS targets without living in Xcode for every iteration.

@acoyfellow introduces capa, a Cloudflare-focused tool that turns OpenAPI specs into deployable Worker service bindings. Instead of hand-rolling wrappers for APIs like Stripe, GitHub, Slack, Twilio, or Jira, you can generate a binding and call it directly from your Worker. The catalog already spans thousands of generated methods across more than a dozen capabilities, which makes this a promising bridge between API ecosystems and agent-friendly Cloudflare workflows.

MacStories highlights Monologue Notes as a dictation app that treats your transcripts like usable data instead of trapping them in a mobile UI. The new notes feature ships with an open toolkit, including a CLI, agent skills, and MCP integrations, so you can search, pull, and reuse meeting notes directly inside terminal-based assistants and coding agents.

@colinhacks introduces Pullfrog, an open source CodeRabbit alternative that runs inside GitHub Actions. It can review pull requests, triage issues, take on coding tasks when tagged, and works with your own model keys or its at-cost router. The pitch is practical: transparent per-run pricing, repo-local execution, and enough built-in tooling to make agent automation feel native to GitHub.

@_ashleypeacock highlights Cloudflare's new automatic tracing across Worker-to-Worker and Worker-to-Durable Object calls, which finally makes distributed traces feel like a single trace instead of a scavenger hunt. External trace-context propagation is next, so third-party services should start showing up in the same end-to-end view too.

@meta_alchemist points out that Telegram now supports streaming responses for agents, making chat feel much closer to using Codex or Claude Code in a terminal. Real-time output matters more than it sounds, because it turns agents from turn-based bots into something you can actually work alongside.

@sebastienlorber highlights a nice upgrade to responsive images with sizes="auto": lazy-loaded img elements can now let the browser infer the right srcset candidate without hand-tuned breakpoints. With support now landing across the latest browsers, responsive image markup gets a little simpler and more robust.

@QingQ77 points to StemDeck, a local-only stem separation app that takes a YouTube link, splits the track into vocals, drums, bass, guitar, piano, and other with Demucs, then gives you a DAW-style multitrack mixer in the browser. A nice open alternative to cloud tools like Moises if you want practice, remix, or transcription workflows without uploads or accounts.

@dok2001 points to Cloudflare MCP server portals, a way for teams to control and organize MCP servers in one place through Cloudflare Access. Paired with Code Mode, the pitch is practical agent infrastructure: central governance, less context bloat, and a cleaner path to sharing internal tools across a team.

@FredKSchott introduces Flue, a TypeScript framework for building headless, programmable agents around a built-in harness. The pitch is Claude Code style ergonomics without the baked-in interactive assumptions: you get a harness-native runtime you can script, embed, and ship inside your own tooling.

@Baconbrix previews serve-sim, a tiny tool that lets you open Apple Simulators in your browser or favorite agent workflow with npx serve-sim. It is a small but very practical bridge for AI-assisted mobile development, especially when you want agents or remote tools to inspect simulator output without living inside Xcode.