Recently updated open-source projects

feature-ux-analyze frames a product feature from a brief or from scratch, gathers competitor interfaces and flows, and turns the results into a node map with P0–P3 insights. It emphasizes findings that remain useful beyond a screenshot, verifies sources, and can inspect video with yt-dlp and ffmpeg. The result is a structured competitive UX teardown rather than an unranked folder of captures.

Owl Browser gives Claude Code a browser MCP server and an observe-and-act skill that renders pages as compact OwlMark text with stable handles. The agent navigates, clicks, and types by handle rather than guessing from screenshots or pixel coordinates. A running Owl Browser instance, endpoint and token environment variables, and npx are required before the plugin can operate.

sw-guardrails bundles a reviewer for Speechworks copy and code plus an outreach-message writer for people who stutter, speech-language pathologists, groups, NGOs, and partners. The checks cover cure claims, fluency-as-success framing, shame, audience language, metrics, events, prompts, and embedded UI text, returning severity, rationale, and a rewrite. The project emphasizes lived experience and honest product boundaries, remains under active development, and is communication guidance rather than clinical advice.

palamedes combines multi-agent research prompts, an agent-loadable skill, calibration scenarios, a browser UI, and offline report templates around a strict evidence discipline. Load-bearing claims need retrievable source text and verbatim quotes, while missing evidence is marked instead of filled from intuition; agreement among agents does not count as independent support. Stakes determine the depth of the workflow, and the browser UI uses the user's own API key.

codex-tui-imagegen drives Codex’s interactive terminal UI inside a tmux pane, waits for the generated PNG to appear in the local image directory, and copies it to the path you specify. It is designed for cases where headless execution cannot access image generation or fails to expose a usable file path. The workflow uses an existing Codex/ChatGPT plan instead of a separate image API, but needs an interactive local session.

This macOS-only toolkit controls Apple Music through natural language, a command-line interface, or an interactive terminal UI, including playback, volume, multi-room AirPlay, playlists, and equalizer controls. Basic playback and speaker routing need no developer account; catalog search, library management, discovery, and API-based playlist operations require an Apple Developer account and MusicKit setup.

youtube-to-text accepts a YouTube URL or local file, extracts audio, and transcribes it with local faster-whisper or ElevenLabs Scribe v2. It outputs paragraph-segmented text, optional SRT subtitles, and an optional summary; Claude can clean up English brand and technical terms in Hebrew transcripts. Apple Silicon macOS is the primary target, with local processing by default, while the cloud engine requires an ElevenLabs key.

redesigner captures a website with Playwright or a native app with Maestro, then creates a navigable mock, design analysis, UX findings, and a React-based redesign with Tailwind and Framer Motion. It can export HTML, Figma, and Pencil artifacts after approval. Logins are manual, and capture stays read-only.

cc-passenger-prototype is a project-specific Claude skill for extending the CC ride-hailing passenger prototype. It can add new business pages, create variants, introduce BEM blocks, or maintain the token and base CSS files while preserving the established OKLCH, BEM, and WebView H5 language. It expects the project's source styles to exist and rejects Tailwind rewrites, third-party app imitation, and direct Vue, React, or UniApp output.

Give a screenshot to a compatible coding agent and the skill builds semantic HTML, then repeatedly renders it in real Chrome, compares the result, and refines layout, spacing, color, and typography. The final deliverable is one self-contained file with inline CSS and JavaScript, plus verified controls where the design implies interaction. Its rendering and verification scripts require a local Chrome installation and Node dependencies.

repo-proof runs a repository’s quickstart inside a sandbox, fixes documentation when commands are wrong, and verifies the result again before capturing a real demo. It refuses to hide an actual failure behind polished copy, making it useful for maintainers who want to demonstrate that a Claude Code skill or open-source project really works. The process executes project commands in an isolated environment.

This pack gives AI coding agents four focused workflows for TikTok, Reels, and YouTube Shorts. Use it to design a stronger opening and pacing, create word-timed highlighted captions from a transcript, build a frame-accurate countdown that lands on zero, or generate branded lower thirds across a speaker roster. It installs through the skills CLI or as a Claude Code plugin and is intended for constructing video templates and motion components.

Product UI Design helps build dashboards, SaaS screens, settings, tables, forms, and admin tools with a restrained product register. It asks the agent to observe real products, separate functional restraint from expressive styling, and scan for common AI-looking choices before delivery. It is meant for product surfaces rather than marketing pages; an optional script checks machine-detectable tells.

Query Japan’s official e-Gov Law API from Claude Code and other skill-compatible agents. The skill finds laws by name, retrieves specific articles, searches full text, and looks up amendment history or the text in force on a past date. It needs no API key, but it relies on bash and curl and does not cover treaties, ministry notices, guidance, or local ordinances.

Query Japan's post-1947 National Diet records by speaker, keyword, party, period, or session, then retrieve meeting lists, individual speeches, or full transcripts. The skill supports policy research, bill tracking, media fact-checking, and academic work without authentication. It still needs an HTTPS fetch tool, and Imperial Diet records before 1947 are outside its scope.

Hyperloop asks you to define the destination and verification criteria, then loops through planning, building, checking, review, and repair until the criteria pass. It scans the project to recognize its stack and can account for common frontend, backend, database, hosting, and auth setups. You still own the acceptance checks and final review; the skill automates the loop rather than promising correctness by itself.

presentation-evaluator scores a talk or keynote using Louise Beale’s framework for speaking powerfully in the age of AI. It checks audience fit, the opening hook, message focus, one clear intent, spoken language, and vivid examples, then asks what should be left out. The result is a one-shot evaluation for sharpening the human layer of a presentation, not a rehearsal loop or slide generator.

solana-incident-response-skill structures a live Solana security response around detection, containment, investigation, fund tracing, and the remaining response tracks. It prioritizes reversible, authority-gated actions such as pausing systems, rotating authorities, locking multisigs, and invalidating pre-signed transactions. Investigators can replay an exploit on a Surfpool mainnet fork to separate a code bug from key compromise. It provides operational support, not legal advice or a guarantee of recovery, and named human authority must approve fund-moving or program-changing actions.