GitHub Trending Today #56: learnharness, tstack, iFixAi, hallmark, project-maya, neuron-explorer Transcript, AI Summary & Key Points

AARE is an open-source Android application that reimplements the phone side of Android Auto without Google Play Services. It communicates with a car head unit using Android Auto's projection protocol, renders the car interface on the phone, and hosts AndroidX Car App Library navigation apps, media services, messaging notifications, voice assistants, phone-screen mirroring, sensors, audio, and instrument-cluster output. It supports USB and wireless connections, includes a desktop head unit for testing, and can use screen capture, Shizuku, or a privileged installation for phone audio routing. The project is early software with limited real-car testing and certificate-compatibility constraints; it is distributed as an APK and desktop head unit under the GPL-3.0-or-later license.

A searchable research archive and chronological dossier about Jesus of Nazareth, covering New Testament, gnostic and extracanonical gospels, church fathers, ancient witnesses, Roman legal documents, archaeology, and related sources. Its site organizes the dossier, source texts, a gospel registry, and citation cards with cross-links between 238 references and the documents they support. Python build tools extract references from the dossier PDF, fetch and normalize sources, generate the searchable site, verify mirrored text, and record source-link matches; the repository also documents public-domain and linked copyrighted materials.

ascii-reveal is a React loading and content-reveal library that makes text, images, videos, canvases, and cards decode through animated character grids while preserving their real DOM position and avoiding layout shift. Its AsciiReveal component draws the scramble and reveal on a canvas layer above the content; it supports loading transitions, selectable presets such as decode, dither, terminal, and blueprint, configurable wave origins and timing, lazy viewport reveals, mascots, replay controls, and reduced-motion fallbacks. AsciiSkeleton provides server-renderable HTML and CSS placeholders that paint before JavaScript, with optional canvas-based shimmer animation. A core API supports use without React. The repository reports a small, dependency-free distribution and is licensed under MIT.

Beautiful Rust is a collection of Agent Skills and a Clippy configuration that teaches coding agents to write, review, port, and enforce idiomatic Rust. Its three skills cover Rust best practices, prioritized reviews of files, crates, or diffs, and ports from C, C++, Python, Go, or JavaScript to Rust; the porting workflow ports tests first and records deliberate API differences. The included lints configuration checks machine-enforceable rules such as documented errors and panics, justified unsafe blocks, prohibited unwrap calls outside tests, lossy casts, and leftover debugging statements. The guidance is distilled from the Google Rust Style Guide, Rust API Guidelines, Effective Rust, Rust Design Patterns, and Microsoft's Pragmatic Rust Guidelines. It can be installed as a Claude Code plugin or through skill directories supported by Codex, Copilot, Cursor, Gemini CLI, and other agents that read the Agent Skills format.

Correlation is a research project and software suite for testing whether stock correlations predict later market movements. Its Rust engine computes all-pairs and lagged correlations using INT8 matrix multiplication through cuBLASLt on NVIDIA GPUs or AVX-VNNI on CPUs; the same engine can process other large collections of time series. A command-line tool and eframe/egui desktop interface provide census, holdout, forecasting, portfolio, diversification, hedging, grouping, timeline, and step-by-step correlation-explanation workflows. The repository also includes Python machine-learning experiments, a grader that compares strategies with turnover-matched random accounts, walk-forward testing with held-out data, datasets, and documented experiment results. Its reported tests found that lagged correlations did not reliably predict stock movements after trading costs, while same-day co-movement was useful for portfolio-risk analysis. The software is released under the GNU General Public License version 3 or later; included market and external-series data remain subject to their own source terms.

An open-source set of agent skills by Enes Ozturk that converts UI components into monochrome, one-page technical drawings modeled on Apple's dimensional drawings. Given a source such as a repository file, Storybook story, URL, or screenshot, the drawing skill produces a title block, plan view, exploded layers, detail views, viewport placement, and variable tables covering architecture such as paddings, gaps, corner radii, and layer relationships; unknown values remain marked as [TBD], and content and branding are omitted. A companion motion skill animates a finished sheet as a roughly 10-second drawing sequence, with a Claude Motion workflow and a headless Playwright and FFmpeg renderer requiring Node 18 or later. The repository is distributed under the MIT license and can be installed as Claude or other agent skills.

Enzyme is a local-first compile step and command-line tool for knowledge bases, developed by byenzyme. It compiles Markdown notes and other sources into a local index by sampling context over time, using tags, wikilinks, folders, and dated records to generate thematic questions called catalysts. These catalysts are embedded and used at query time to find relevant documents, giving AI agents an orienting map of a knowledge base without generating full summaries for every search. The tool supports Markdown folders, Apple Notes, Are.na channels, and experimental SQLite message archives, with commands for compiling settings, initializing and refreshing an index, inspecting the resulting concept map, and running hybrid searches that return source passages together with the catalyst connecting them to the query. It can use hosted, local, or OpenAI-compatible models, and can index and search without catalyst generation when no model is configured. Enzyme is licensed under the Apache License 2.0.

ESP32 Labs is a React-based interactive bench guide for learning the ESP32-S3 microcontroller. It provides 32 hands-on labs with live pinout views, wiring instructions, flash commands, browser-based simulators, a 3D DevKitC-1 board, a bench log, and a code reader for the associated PlatformIO firmware. The code reader offers guided tours, execution stories, line explanations, and a phrasebook for the C++ used in each lab. It also includes a cited, maker-focused ESP32-S3 datasheet reader covering pins, peripherals, memory, clocks, and power modes. The guide supports English, French, Spanish, Italian, German, Polish, Turkish, Simplified and Traditional Chinese, Japanese, and Arabic, with progress stored locally in the browser. It is a Vite-built React application deployable to Vercel and is an independent learning project, not affiliated with Espressif.

An evidence-first genealogy skill for AI agents such as Codex and Claude Code. It supports family-history research across the former Soviet Union and Russian Empire, covering church books, civil, military, deportation, repression and unindexed archive records across all 15 former Soviet republics. The skill resolves historical jurisdictions for each event, generates name-variant and transliteration matrices, prioritizes original records over catalogs, indexes, trees and forums, and maintains registers of people, artifacts, claims, relationships and searches. Its evidence protocol separates records from interpretation, assigns statuses such as hypothesis, corroborated, contradicted and unresolved, groups copies of the same record to prevent false corroboration, and keeps namesakes separate until distinguishing evidence appears. An audit script checks register integrity, citations and corroboration before reporting. It also provides workflows for targeted FamilySearch research and World War I source searches, archive correspondence, multilingual source queries and final reports that document relationships, contradictions, failed searches and next steps. The repository includes Python scripts using the standard library, with Chromium-based browser support required for PDF generation. It is distributed under the MIT license and is maintained by Denis Shiryaev.

farol is a free, open-source CLI and GitHub Action for assessing pull-request merge risk and showing where changed code reaches. It statically compares the base and head trees using the TypeScript compiler API, go/types, and Python AST parsing, builds declaration and call graphs, and traces changed declarations to affected pages, API operations, RPC methods, jobs, scripts, and other entry points without executing the analyzed code. It combines reach, diff coverage, lint and security findings, dependency changes, and contract breaks into a 0–100 risk score with fast-track, quick-review, or careful-review guidance, then ranks files and identifies checks for the reviewer. It supports TypeScript/React, JavaScript and TypeScript backends, Go, and Python frameworks and tools, and can produce Markdown comments, JSON, or single-file HTML reports; optional Datadog traffic and error data can add operational context. farol runs locally or in CI with no account or hosted service, never sends analyzed code elsewhere, never blocks or approves merges, and is distributed under the MIT license.

gdirect is a Rust command-line tool for browsing, downloading, uploading, and mounting Google Drive. It can export Google Docs files to formats such as PDF, resume interrupted transfers, verify completed downloads against Drive checksums, and manage multiple Google accounts. For transfers, gdirect scans reachable Google front ends, tests them with TLS, Drive API, and download checks, and routes requests through verified endpoints. It supports multi-connection downloads using pinned libaria2, automatic retries between endpoints, and optional use of a relay or Google-only routing. Mounting uses rclone, with NFS on macOS and FUSE3 on Linux; local writes are cached and uploaded in the background. The project provides prebuilt binaries for macOS, Linux, and Windows, plus source-build instructions.

Hallmark is an open-source design skill for Claude Code, Cursor, and Codex that applies anti-AI-slop rules to generated interfaces. It selects a macrostructure and one of twenty-one themes for a brief, applies its design rule set, runs fifty-seven anti-pattern gates and a pre-emit self-critique, and returns self-contained HTML and CSS. Its four operations build new UI, audit existing code without editing it, redesign an interface while retaining its copy, information architecture, and brand, or study a screenshot or URL to extract macrostructure, type pairing, and a color anchor; the study operation can also emit a portable design.md file. Custom briefs can receive a made-to-measure palette, typography, and layout rather than a catalog theme. Hallmark is made by Together AI and distributed under the MIT license.

Hob is a free, open-source desktop application from Frigade for running open-weight AI models locally on macOS, Windows, and Linux. Built with Electron and Llama.cpp, it provides an alternative to hosted assistants such as Claude and ChatGPT and can fill out PDFs, work through documents, analyze spreadsheets, and create charts. Hob selects an open-weight model suited to the user's computer and sets it up for local, private, offline use. It is distributed under the MIT license.

Horizon is an unofficial, open-source reimplementation of Minecraft: Java Edition in Rust. It runs as a native game through the Horizon launcher, reads worlds in Minecraft's format, reproduces vanilla world generation and game rules, and requires the user's own legally obtained copy of Minecraft: Java Edition; it contains no Minecraft code, assets, or data. The project checks behavior against vanilla with differential harnesses and supports single-player worlds, resource and data packs, controller input, optional graphics and navigation features, offline-mode multiplayer, and a Rust-to-WebAssembly mod system. Java mod JARs are not supported. The launcher installs and updates the game, can manage Horizon servers and modpack profiles, and the distribution includes horizon-ctl, a command-line and MCP interface for controlling the game programmatically.

iFixAi is an open-source tool for independently auditing AI agents against business, governance, safety, and operational requirements. It runs a diagnostic through a CLI, an agent plugin or skill, or an HTTP/provider adapter; the agent under test is evaluated with inspections covering tool authorization, privilege escalation, prompt injection, policy violations, private-data handling, uncertainty, failures, instruction following, and related risks. Runs produce JSON and Markdown reports with a scorecard based on five core pillars: fabrication, manipulation, deception, unpredictability, and opacity. The repository includes 60 inspections, configurable suites, self- or independently judged evaluation, fixture authoring, and support for testing a deployed agent or a bare model. The engine is free and distributed under the Apache 2.0 license; iFixAi Pro is a separate auditing service with additional inspections, reports, and an audit dashboard.

Lantern is a zero-dependency TypeScript library for animated, step-by-step explainers of code changes. It uses a JavaScript port of manim's core model to represent scenes as trees of point buffers and renders them as SVG in browsers, Cloudflare Workers, or Node. Scenes define asynchronous construction methods with named steps; each step can expose state, and the same scene can produce live DOM output, SVG frames, text representations, structured traces, or motion-verifier samples. It includes Mermaid flowchart parsing and JSON document scenes, supports deep links to individual scenes and steps, and uses fixed-timestep playback so seeking can reproduce the same output as a full run. The repository is written in TypeScript and has no runtime dependencies; it is not currently published on npm.

LearnHarness is an open-source desktop application that uses AI agents to teach difficult subjects from the learner’s current understanding to a stated goal. It measures understanding through answers, confidence, and explanations; builds a dependency plan; teaches through prediction, explanation, and comprehension checks; and uses adaptive practice to test whether knowledge can be recalled without being prompted. A persistent local knowledge graph stores propositions, reasoning chains, and links to source sessions so later teaching and practice can build on previous work. The app can research questions, create visual explanations, support dictation, and use configured model providers. It runs on macOS, Windows, and Linux, uses the learner’s provider API keys rather than a LearnHarness subscription, stores chats and learning data locally, and is released under the AGPL-3.0 license.

LinuxRunner is an open-source iOS and iPadOS application for running ARM64 Linux userspace, terminal programs, and desktop applications on iPhone and iPad. It supports Debian GNU/Linux ARM64, Arch Linux ARM, and Alpine Linux ARM64, with package managers, Linux accounts, X11 desktop sessions, graphical applications, touch and hardware-keyboard input, and optional file and network integration. Its executable loader maps ARM64 Linux ELF programs and libraries, while a custom interpreter executes supported guest instructions through precompiled host routines. A Linux compatibility layer implements system calls, processes, threads, signals, memory mappings, filesystem metadata, and Linux sockets through platform networking APIs. X11 applications use Xvfb and x11vnc with a native desktop viewer; Mesa provides software OpenGL and an experimental Metal bridge supports selected rendering paths. LinuxRunner uses userspace compatibility rather than a virtual machine or hardware virtualization, does not boot a Linux kernel, and does not require JIT in the release build. The application is distributed as an IPA for compatible sideloading methods and does not require a jailbreak. It supports ARM64 devices with iOS or iPadOS 15.0 deployment targets, although practical workloads depend on available memory and storage. x86 and x86-64 applications, Linux kernel boot, Wayland sessions, systemd as init, kernel modules, and conventional Linux containers are unsupported; several desktop and graphics paths remain experimental. LinuxRunner is maintained by Leminos and is covered primarily by GPL-3.0, with upstream and third-party components retaining their stated licenses.

Minilytics is an open-source, self-hosted web analytics platform for tracking visitors, revenue, conversions, referrers, campaigns, pages, outbound clicks, button clicks, goals, funnels, and visitor journeys. It runs on Cloudflare Workers and D1, with a browser tracker and a React dashboard; each deployment stores its data in the deployer's Cloudflare account and can serve multiple websites. Revenue attribution assigns a payment to the visitor's last visit from a non-direct source. It supports Dodo Payments and Stripe webhooks, payment-link and checkout metadata, return-page matching, customer and email fallbacks, and a generic payments API; refunds are subtracted from revenue. The dashboard also provides live visitor maps, source filtering, date ranges, and customer timelines. The project can be self-hosted on Cloudflare's free plan using Node.js and Wrangler, or configured as a hosted product with Google sign-in and Dodo Payments. It is released under the MIT license.

MotionSites Web Builder is a free Claude skill for creating cinematic landing pages. Claude studies related example designs, selects media from a shelf of hosted videos, 3D cut-outs, and images, writes a detailed build specification, generates the page, then uses screenshots and a QA script to check assets, mobile overflow, fonts, and video playback before applying fixes. The skill includes example prompts, technique recipes for effects such as seamless video loops, liquid glass, word-by-word blur headlines, mouse-scrubbed video, depth carousels, scroll stories, and Three.js models, plus design rules and a specification template. It can be installed in Claude.ai, Claude Desktop, or Claude Code, and can produce prompts for other website-building tools. Its instructions and code are licensed under MIT; the example prompts and media remain under MotionSites copyright.

An open-source interpretability tool for Qwen3.5-4B. It pushes each MLP neuron in a layer up and down across selected prompts, retains the resulting rollouts, and presents them in a three-dimensional model view. The explorer includes steer and logit lenses, neuron rankings based on word-cluster mass against random-direction controls, activation-based search, steered chat, neuron connections, and UMAP views of rollout embeddings. It can load Qwen3.5-4B locally on MPS, CUDA, or CPU, and includes sweep scripts for collecting activation statistics, steering runs, token log probabilities, lens data, depth reports, and neuron connections. Its optional Claude panel uses the explorer as an MCP server for inspecting neurons and running steering, activation, ablation, and inference tools.

OmarchyTV is an unofficial Omarchy-styled launcher ecosystem for Android TV and Google TV, with experimental editions for LG webOS, Android mobile, Linux, and macOS. Its TV launcher uses five workspaces, Hyprland-style windows, a Waybar-style top bar, fuzzy search, configurable app panels, Live TV and HDMI access, bundled themes and wallpapers, and ASCII screensavers. The companion PC/macOS plugin pairs over an encrypted LAN connection, synchronizes themes and wallpapers, forwards PC telemetry, mirrors the desktop, and exposes a TV-control MCP server. A built-in Omarchy TUI assistant can control local TV functions, launch installed apps, and bridge open-ended questions to authenticated desktop AI command-line tools; an Android companion can install and update the TV launcher wirelessly. The project is beta software, is distributed under the MIT license, and is not affiliated with or endorsed by Omarchy.

OpenIDEA is a cross-platform desktop IDE with a Rust core and an IntelliJ IDEA–style keymap, tool windows, and search-everywhere workflow. It combines an editor, code insight, debugger, Git integration, terminal, database client, and HTTP client for more than 60 languages, and builds for macOS, Windows, and Linux. The project uses Tauri, is licensed under Apache 2.0, and is not affiliated with JetBrains.

pngdeer is a lossy PNG compressor written in Rust, available as a command-line tool and an in-browser WebAssembly application. It searches palette sizes from 2 to 256 colors, tests dithering and other palette-generation options, and scores each result with the SSIMULACRA2 perceptual metric. It keeps the smallest file that reaches its target quality, selecting the quality automatically or accepting a fixed target. The command-line and browser versions use the same Rust code and produce identical files; browser compression runs locally without uploading images. pngdeer is free software released under the MIT license.

Project Maya is an open-source local inference system for running large mixture-of-experts language models on personal hardware. Created and led by Peasant Smith and co-maintained with needmorevram, it provides model-specific engines for GLM-5.3-Flash, DeepSeek-V4.1-Flash-1017, and experimental MiniMax-M3.1-Flash, with support for NVIDIA or AMD GPUs, Linux, and experimental Windows configurations. Maya places frequently used experts in GPU memory, keeps additional experts in RAM, and reads the remainder from an NVMe SSD, moving them as inference proceeds. Its setup measures the available GPUs, CPU, RAM, and storage and tunes the configuration; layers can be split across multiple GPUs, and optional drafter, quantization, vision, KV-cache streaming, and conversation-checkpoint features are selected according to the model and hardware. The project includes a browser chat interface and hardware monitor, an OpenAI-compatible API, an Anthropic-compatible API, tool calls, image inputs, thinking-level controls, concurrent requests, and a Responses API with streaming and function-tool support. It is free and distributed under the MIT License; the repository notes that some optional model components have separate licenses or non-commercial restrictions.

Raven is a Linux runtime for running Windows applications without a virtual machine or Wine. It runs the Windows program code directly on the CPU inside a sandbox and answers its system calls by implementing Windows-kernel and service behavior, including processes and threads, files, the registry, windows, and input. Graphics are routed through Wayland and Vulkan, with DXVK and vkd3d-proton providing Direct3D support. Applications are installed with their normal Windows installers into separate prefixes, each with its own C: drive and registry. Raven supports both 32-bit and 64-bit programs, Windows Installer, NSIS and Inno Setup packages, selected .NET Framework versions, and Direct3D 8 through 12 through its graphics components. It is currently built from source for x86_64 Linux using CMake, C++20 and Python 3.11 or newer. Raven's Linux runtime is licensed under GPL-3.0-or-later, while its guest adapters and imported components retain their stated licenses.

rederive is a skill for Claude Code and other coding agents that teaches a codebase by having the user re-derive its design decisions from first principles rather than receiving explanations. It asks questions about why the project exists and how the user would solve its problems, while a background agent maps the repository by reading code, import relationships, and design-changing commits. The resulting map contains roughly 5 to 10 problems ordered by how the system would be built and is stored in `.rederive/map.md`; the session records difficulties and independently worked-out solutions in `.rederive/log.md`. When the user is stuck, rederive narrows the problem to several approaches and can offer a sketch after two consecutive incorrect answers. It can be installed with `npx skills add EricJujianZou/rederive`, used with `/rederive` on a local or GitHub repository, and also installed as a Claude Code plugin. Its local files are excluded from the repository's commits, it sends no telemetry, and the project is licensed under MIT.

StudyAI is a local-first study platform for university lectures. It imports PowerPoint files and extracts text, tables, charts, SmartArt, speaker notes, and equations, displaying the original slide beside the extracted content and providing search across lectures; images whose text cannot be extracted are flagged for manual entry. It supports FSRS spaced-repetition flashcards, notes, practice exams with marking points, progress tracking, and ZIP backups. An optional AI tutor runs through a user-managed Cloudflare Worker with a Google Gemini free-tier key, answers from the user's slides, verifies cited quotations against the source slides, labels information beyond the slides, and checks generated exam questions before approval. Lecture and study data remain in the browser by default, the app requires no account or third-party requests for its non-AI mode, and the repository is licensed under MIT.

SuperGhidra is a Python reverse-engineering suite built around Ghidra. It recovers MSVC RTTI, MFC CRuntimeClass hierarchies, virtual tables, and Serialize() archive layouts from PE binaries; loads ELF, PE, and Mach-O formats; matches functions and shifted struct fields across binary versions; rewrites Ghidra decompiler output into compilable C; and provides a Z3-based P-code equivalence checker. Its persistent read-only Ghidra daemon keeps a JVM session warm, uses an SQLite cache for queries, and records mutations in a JSON overlay and transaction journal rather than writing directly to locked .rep projects. Standalone Python, NumPy, WGPU, and Z3 modules handle binary analysis and selected accelerated scans without Ghidra, while an optional SuperGhidraMCP server exposes reverse-engineering tools to agents such as Claude Desktop, Cursor, and Antigravity. The standalone zero-JVM decompiler is explicitly experimental and supports only a bounded instruction subset; the SMT checker returns UNKNOWN for unsupported operations, floating-point code, and overly long loops. SuperGhidra is distributed through PyPI and source installation under the GNU AGPL-3.0-or-later, with an optional commercial license.

Tellyscope is an Android app written in Kotlin that manages Android TV and Google TV devices, with stated support for NVIDIA Shield and Fire TV, from a phone over Wi-Fi. It uses ADB and supports network debugging and Android 11+ wireless debugging with pairing codes to debloat devices, install APKs and app bundles, transfer files, control the TV remotely, change launchers, block ads with Private DNS, adjust animations, clear caches, and inspect CPU, RAM, storage, temperature, processes, and network use. Debloating disables built-in apps rather than uninstalling them, shows the affected apps first, records changes in per-TV history, protects essential apps, and provides undo and restore functions; apps installed by the user can be uninstalled. The app requires Android 8.0 or newer on the phone, does not require root or a PC, and states that it has no ads, tracking, or accounts. It is distributed as an APK through GitHub Releases and is licensed under the GNU GPLv3.

tstack is a collection of agent skills for Claude Code and Codex. Its skills automate software-development workflows such as proving, reviewing, filing, monitoring, rebasing, and merging pull requests; the council skill obtains reviews from three models from different providers and loops until they agree, while sol-loop repeatedly reviews a change with GPT Sol. Skills can call other skills, and the repository includes workflows such as full-send, full-stop, takeover, babysit-pr, modernize, file-pr, and automatic pull-request comment formatting. It is installed by copying the repository's skills into the Claude Code or Codex skills directory. An authenticated GitHub CLI and Python 3 are required for some workflows, while council and sol-loop require access to models from other providers. The repository is licensed under MIT.

An open-source toolkit for slimming Android TV and Google TV devices over a local network. It connects through Android Debug Bridge (ADB), diagnoses the TV, saves its app and settings state, and disables selected advertising, telemetry, promotional, and other system services using brand-specific package lists; the changes are reversible and the toolkit does not uninstall protected apps. It can also install Zen Home, a minimalist launcher with configurable titles, four-, six-, or eight-app layouts, favorite-app docks, photos, themes, weather, and optional replacement of the original home screen. The toolkit provides Windows, macOS, and Linux applications or scripts, downloads ADB on first launch, checks the target TV's serial number before actions, and includes a restore option. The kit and Zen Home source are released under the MIT license; the repository describes it as a personal, unsupported project with no warranty.

webframe is a cross-platform Go command-line tool and library that runs a web page like a program in a hidden system webview. It loads URLs or local files, sends console output to stdout and page exceptions and resource-load failures to stderr, and can evaluate JavaScript from standard input, waiting for promise results. It uses the operating system's WebKit or WebView2 implementation rather than bundling a browser, and can run browser-only code such as WebAssembly, WebGPU, Web Audio, and DOM libraries from scripts, CI, cron jobs, or long-lived processes. The CLI supports headless operation, optional visible windows and web inspection, environment-variable globals, Docker, and WebSocket or standard-input interaction; the Go API exposes navigation, evaluation, event callbacks, and lifecycle control. It is distributed as platform-specific releases or built with Go, and is licensed under the MIT license.

whatsmyagentdoing is a local dashboard and Claude Code/Codex skill for monitoring Claude Code, Codex, Gemini CLI, and OMP agents on a machine. Invoked with `/myoffice`, it discovers the agents' JSONL session logs, tails them read-only, converts recent events into statuses, and sends updates to a browser over Server-Sent Events. The interface presents active agents in an isometric office, with 2D, 3D, and list views; it shows states such as thinking, running a tool, writing a reply, waiting for the user, done, interrupted, and error, and can display the current tool, resume commands, notifications, filters, downloadable scenes, and an always-on-top pop-out where supported. It requires Node.js 18 or newer, has no runtime dependencies or API keys, binds only to localhost, and is distributed under the MIT license.