GitHub Trending Weekly #45: comet, blobatar, cumora, herdr, barehands, ip-as-logo-skill, Aura, md2hd Transcript, AI Summary & Key Points

An open-source Python tool that extracts local chat histories from AI coding assistants and converts them into a normalized JSONL format for backup, personal analytics, or fine-tuning. It auto-discovers conversation data in the storage formats used by Claude Code, Codex CLI, Cursor, Windsurf, Trae, Continue, Gemini CLI, OpenCode, Cline/Roo Code, and Aider. Extracted records can include user and assistant messages, code context, diffs or suggested edits, tool calls and results, timestamps, session IDs, project paths, and model names when those sources store them. It handles JSONL, JSON, SQLite, Markdown, and heuristic undocumented schemas, searches macOS, Linux, and Windows storage conventions, and runs from the command line using only Python's standard library, requiring Python 3.9 or newer.

AI Design Skills is an MIT-licensed collection of Markdown design skills for AI coding tools such as Claude Code, Cursor, Codex, Windsurf, and Cline. Its skills provide coding agents with guidance for landing-page strategy, page structure, conversion copy, typography, spacing, visual systems, border radius, and motion. Each skill is stored as a self-contained folder with a SKILL.md file that can be copied into an agent's project-rules directory, added to a system prompt, or uploaded to a Claude.ai project. The repository currently documents a landing-page-design skill that asks intake questions, builds the page structure and copy, and defines the visual system; its design values are grouped in an editable section so users can change fonts, colors, spacing, and motion. The collection can be forked and used, modified, distributed, or used in commercial work under the MIT license.

APEX is an open, verification-first RTL design for an LLM-inference tile developed with Sigmantic AI. It implements one transformer decoder layer—including attention, softmax, RMSNorm, RoPE, SwiGLU, residual processing, and KV-cache compression—and runs Qwen2.5-0.5B through the verified pipeline on FPGA hardware. The KV-cache codec sits in the datapath: cached keys and values are compressed as they are produced and decompressed when consumed, while an importance unit allocates precision according to which parts of the context matter. Each RTL block is checked against an executable NumPy golden model with bit-exact comparisons and mutation-tested testbenches. The repository covers the inference tile rather than a complete system; DRAM control, PCIe, and NoC are out of scope.

arc-code is a minimal harness for testing whether a general coding agent can learn unfamiliar ARC-AGI-3 games during play. It runs a headless coding agent with a shell, filesystem, and action command inside an internet-isolated sandbox; the harness provides no solver, planner, world model, grid tooling, fine-tuning, MCP servers, subagents, plugins, or hooks. The agent interacts with each game only through `act.py`, which executes actions and appends resulting boards and actions to a log. It uses the shell and approved file tools to parse that log, record findings, and build task-specific machinery such as parsers, rule models, simulators, and search procedures, which are discarded when the game ends. `run.py` launches and records one session per game, while the `rig/` directory supplies sandboxing, brokering, auditing, scoring, and export infrastructure. The repository includes the harness and six complete session records.

Aura is an open-source web gallery and customizer for ambient, layered gradient backgrounds built with CSS blend modes. It provides live previews and a fullscreen viewer, then lets users add, remove, and reorder layers while adjusting blend modes, blur, opacity, base color, and an optional grain overlay. Gradients can be exported as CSS, Tailwind, CSS variables, or CSS-in-JS, and downloaded as PNG or SVG images; the tool also generates prompts for recreating a gradient with ChatGPT, Claude, or Copilot. The repository is MIT-licensed and provides a local development workflow using pnpm.

barehands is an open-source, webcam-powered hand-tracked interface for AI assistants. It runs in Chrome with a webcam and overlays notes, images, and 3D models as cards over the camera feed; users manipulate them with gestures such as pinching, throwing, stretching, force-pulling, and two-finger dragging. The interface uses a local Python server and loads Google MediaPipe for hand tracking and three.js for 3D rendering. Markdown folders, including Obsidian vaults, can serve as note sources, while configured media folders provide images, transparent props, and 3D models. An AI assistant can control the interface through local files and shell scripts: state files drive the on-screen ring's status, and the board protocol presents content on the stage. The repository states that it can be used, modified, and built on commercially within a business, with redistributed versions remaining under the same license.

blobatar is a dependency-free JavaScript library for generating deterministic geometric SVG avatars from strings such as usernames, display names, emails, handles, IDs, or repository names. The same input string always produces the same blobatar, which can represent a user, bot, team, or other named entity. It provides a core JavaScript interface and framework adapters for React, Vue, Svelte, Solid, Preact, React Native, and Expo. Outputs can be used as SVG markup or data URIs, with controls for visual traits, palettes, expressions, and optional animation. The package is distributed through npm-compatible package managers and is reported by its repository to be approximately 4.4 KB gzipped.

Core Framework is a visual CSS framework and design-token builder for the web, WordPress, and Figma, developed by CoreBunch. It provides a visual editor for semantic colors, light and dark values, generated shades, fluid typography and spacing scales, layouts, reusable selectors, components, utility classes, and design primitives such as radii, borders, shadows, opacity, filters, and transforms. The system calculates responsive type and spacing scales from minimum and maximum values, scale ratios, and naming conventions, then generates standards-based production CSS. Projects can be previewed in the editor and their readable or minified stylesheets can be copied, exported, or downloaded, or saved as portable .core files. The same open-source codebase supports a standalone web app, WordPress, Figma, Gutenberg, Bricks, and Oxygen integrations. The editor is local-first, and the web app, plugin, integrations, and builder add-ons are available under the MIT license without paid feature gates.

Cumora is a cross-platform team-chat platform in which AI agents participate alongside human teammates. It provides shared rosters, direct messages, group conversations, a Kanban board, and a calendar; agents can maintain personas and memory, claim work, coordinate, and send and receive email. Agents can run in Cumora's cloud in per-agent Kubernetes pods, using a multi-hop tool-calling loop on the OpenAI Responses API, or through its BYOA mode, where a local daemon connects the platform to agent CLIs such as Claude Code, Codex, Grok Build, Cursor Agent, OpenCode, or pi. The frontend uses React, Vite, TypeScript, and Tailwind across web, desktop, and mobile shells; the backend is a stateless Node service using Express and WebSockets, with PostgreSQL as the source of truth and Redis for pub/sub and presence. Agent coordination uses a freshness gate that holds stale replies for reconsideration, atomic claims on work units, and a triage gate intended to reduce unnecessary large-model calls.

DesktopFly is a macOS desktop companion that places a procedural 3D fruit fly on a transparent, click-through overlay. The fly walks across window edges, grooms, sleeps, flies, and responds to cursor movement, clicks, window changes, keyboard activity, time of day, and thermal state without requiring permissions or entitlements. A separate interactive brain window displays 23,210 FlyWire neuron soma positions and live spikes. Its behavior is driven by a 1 kHz leaky-integrate-and-fire simulation of a 668-neuron circuit containing approximately 19,000 real synaptic connections. Cursor approach is converted into looming input for LC4 and LPLC2 visual neurons; escape is produced only when activity reaches the DNp01/Giant Fiber escape command neuron through the modeled synapses. Other circuit outputs drive walking speed, steering, grooming, backward movement, wing effort, escape maneuvers, arousal, and spontaneous takeoff. The loop also feeds gait rhythm to ascending proprioceptive neurons and fast cursor motion to sensory wind partners. The body geometry and sensory transduction are procedural or modeled, while the downstream connectivity and synaptic strengths come from FlyWire data. The project targets macOS 13 or later and requires the Xcode Command Line Tools. Menu-bar controls provide pausing, brain-window visibility, escape testing, movement between displays, adding or removing flies, and starting a collective startle response; the brain window can stimulate nearby circuit neurons for 400 ms. The code is MIT-licensed, while derived FlyWire data files are CC BY-NC 4.0.

dgit is a Git server and forge for Cloudflare Workers and self-hosted machines using Durable Objects. Each repository is a Durable Object with a private SQLite database that stores repository objects and references, implements Git's smart HTTP protocol for standard Git clients, and provides a cgit-style web interface. Reads are public by default, while pushes require authentication; pushing to a new name creates the repository. The TypeScript implementation includes Git packet-line framing, pack-file parsing and generation, ofs- and ref-delta resolution, commit/tree/tag codecs, SHA-1 streaming, and Myers diffs. Pushed data is retained as the client's original compressed pack file, with an index mapping object IDs to packs, offsets, and delta bases. An optional Cloudflare R2 binding offloads pack bytes and caches full clones, while SQLite retains the index; clones copy stored compressed bytes into outgoing packs, and fetch negotiation supports shallow, incremental, thin-pack, forced-update, and ref-deletion operations. Its web interface includes repository summaries, refs, searchable logs, per-path history, trees, blobs with syntax highlighting, blame, commits and range diffs, format-patch output, archive snapshots, README-based about pages, Atom feeds, and commit-activity statistics. Deployment uses npm and Wrangler, with an optional R2 pack cache and a configured push token. Cloudflare Workers impose memory and CPU limits, so very large histories may require smaller pushes; repositories with millions of objects may be better suited to the project's celld deployment.

Endoplexity is a Chrome MV3 side panel and local bridge for agentic control of the browser a person is already using. It lets Claude or Cursor agent CLIs operate existing tabs and logged-in sessions by clicking, typing, filling forms, navigating, switching tabs, uploading files, and reading pages, without sending requests directly to a model or requiring a separate metered API key. The side panel owns the Chrome DevTools Protocol connection and communicates over an origin-pinned WebSocket with a bridge on localhost; the bridge exposes browser tools over token-gated MCP/HTTP and enforces the safety policy, including approval gates and autonomy modes. Pages are provided to the agent as accessibility-tree snapshots rather than raw HTML, and actions return the resulting page state so subsequent turns can resend only changed lines. Its file-reading tool extracts text from PDFs, DOCX, XLSX, PPTX, CSV, JSON, Markdown, and other text files.

herdr is a terminal multiplexer and persistent runtime for coding agents, developed as a Rust binary. It runs a background server that keeps agent terminals and sessions available across terminal disconnections, SSH reconnections, network loss, lid closure, and machine restarts; users can reattach from another terminal and use tmux-style keyboard controls or mouse interactions to split, move, and manage panes. Each pane is marked working, blocked, or idle, and agents can control herdr through its CLI and socket API to spawn panes, prompt other agents, and wait for an agent that is blocked. It hosts existing tools such as Claude Code, Codex, Cursor, OpenCode, and Grok without wrapping or replacing them, and supports plugins for extending panes and workflows. The project provides installation scripts and package-manager installation, documents remote use and session state, and is licensed under the Apache License 2.0.

HQBase is an open-source shared email workspace for teams that runs in the customer's own Cloudflare account, keeping the application, mail, and Cloudflare credentials in infrastructure the customer controls. It provides shared mailboxes, team access controls, multi-domain configuration, collaborative drafts, and audit history, along with installation, update, backup, and recovery operations. HQBase also includes an OAuth-protected remote MCP server for controlled access by compatible agents. The project is distributed under the GNU Affero General Public License v3.0 only.

icm-architect is a Claude skill that designs processes, ideas, and problems as ICM (Interpretable Context Methodology) workspaces, using folder structure as agent architecture, or restructures an existing folder, repository, or vault. Numbered folders encode sequencing, hierarchy scopes context, and plain Markdown files store state, allowing an agent to orient and act by reading the relevant files. In build mode, it extracts stages, human approval points, and stable versus per-run elements, then scaffolds a workspace using one of six forms: Pipeline, Umbrella, Record library, Knowledge bundle, Context map, or System map. In restructure mode, it audits files as catalog, contract, factory, product, or dead, proposes a migration map for approval, then migrates and validates the result. The walk test checks whether an agent with no prior memory can orient, act, and report status from the workspace alone. It can be installed as a Claude Code skill or uploaded to Claude applications, and is MIT licensed.

IP as Logo Skill is a compact Agent Skill in the open Agent Skills format that guides compatible AI agents in turning a product brief into highly simplified, rounded mascot-character concepts. It gathers product context, proposes three design directions, and after approval generates six independent full-resolution square candidates using one dominant silhouette of roughly four to seven large shapes, two mascot or IP colors, a named solid background color, thick rounded forms, and lower-left or lower-right corner emergence. Its default batch uses two variants per direction with a three-left, three-right composition split rather than a contact sheet. Familiar animals are the default subjects, while objects, machines, fantasy artifacts, and other unusual subjects require a clear product-related reason. The skill provides instructions and prompts rather than an image generator, can be installed with the Agent Skills CLI, and is intended for agents such as Codex, Coze, Doubao, YouMind, Manus, Gemini Apps, and Replit Agent when paired with a supported image model.

jit is a local-first macOS command-line secrets manager from jitpass that moves development credentials from plaintext files such as .env, AWS credentials, shell exports, npm configuration, and MCP configurations into an encrypted vault protected by Touch ID. It rewrites source files with decoys so existing tools can continue working while the real value is released only to an authorized process. It provides credentials through per-process environment injection, native credential protocols such as AWS credential_process and Docker or Git credential helpers, and a named-pipe mount for tools that only read files. The named pipe is a mode-0600 POSIX FIFO: authorized reads receive decrypted values through the kernel pipe buffer without writing them to disk, while ambient readers receive decoys. jit is macOS-only for Apple Silicon and is still in development; the project states that it does not protect credentials after they enter the requesting process's memory.

macOS Harness is an experimental, MIT-licensed, macOS-only Python harness from browser-use that gives an LLM or coding agent raw primitives for controlling a Mac in one persistent Python process, without app-specific tools, recipes, or framework rails. Its six core primitives—`see`, `key`, `type`, `click`, `ax`, and `script`—capture native and Electron app windows, send keyboard and coordinate input directly to an application process, expose Apple Accessibility and Apple Events, execute AppleScript, and draw a click-through pointer without moving the physical cursor. The agent can write missing task logic in ordinary Python during execution. The same process provides access to a real, logged-in Chrome browser through Browser Harness, as well as the local filesystem and shell commands through Python, `Path`, and `subprocess`. It can capture background application windows without bringing them to the foreground and does not activate or raise the target app. The project includes installation and skill-registration workflows, a `doctor` command for checking required macOS permissions, and verification instructions. Anonymous telemetry is enabled by default and records the CLI command category, success, duration, package version, operating system and architecture, and detected agent client; the project states that it does not record prompts, app names, screenshots, UI text, scripts, paths, or window titles, and provides `macos-harness telemetry disable` to turn telemetry off.

md2hd is a local CLI that turns a Markdown file or folder of notes into an interactive relationship map in a browser. YAML front matter becomes nodes, while wikilinks and relationship entries become directed edges; an optional map block defines the map's title, types, and relationship inverses. The map supports searching, type-based views, node-focused navigation, degree-based relationship expansion, dragging and saving positions, and editing the source Markdown through a live code pane. It runs a local server bound to 127.0.0.1, reads changes from disk on refresh, and is described as requiring no account or telemetry. The package also includes a writing-md2hd-maps skill for coding agents that author or convert Markdown maps.

morphnext is a Flutter and Dart library for interruptible, spring-driven vector morphing between bundled Flutter IconData glyphs. It converts icon-font outlines into morph plans, allowing icons to transition between shapes without separate animation assets for each pair. AnimatedMorphIcon provides implicit transitions that continue from the shape currently on screen when the target changes, while MorphIcon accepts an Animation<double> for controlled progress in gestures, scrolling, or timelines. The widgets follow Flutter icon conventions, including theming, visual font controls, directionality, semantic labels, and reduced-animation settings. Fonts and morph plans are loaded lazily and cached with bounded memory; if a font is unavailable, malformed, or unsupported, the library preserves the layout and displays a regular static Flutter icon instead of performing a continuous morph.

NorthCinder is an open-source, local-first MCP server for AI shopping agents. It compares products from user-selected sources against a buyer's criteria, explains rankings and exclusions, shows source facts and unverified details, and normally returns up to three useful options rather than claiming one universal best product. Its research workflow provides separate product and seller guides, creates a research plan, and uses controlled research tools; results remain provisional when sources conflict or do not identify the exact item or seller. Purchasing is a separate human-in-the-loop decision: each checkout requires fresh approval for one exact offer and unit, with the merchant, variant, price, known total, and spending cap signed into a single-use approval. It rejects raw card details, using an opaque payment token for supported automated checkout or handing the buyer a cart link. NorthCinder runs alongside an MCP-capable AI application. Its local setup requires Node.js 20 or later; the init command stores configuration locally and starts the MCP server and search engine in one process. The repository states that there is no hosted NorthCinder account or cloud service, and that order outcomes remain local.

An open-source web application for generating procedural interface sounds live in the browser rather than using prerecorded audio files. Users choose a category, generate variations, audition sounds in UI context, and export selected results as WAV files or JavaScript recipes. Its sound library is hand-curated, while generators draw from and remix the library; curator keep/delete decisions adjust each generator's future output. The project is in early beta, with a local workbench for library curation, recipe editing, calibration, deduplication, and generation. It runs as a Next.js, React, Tailwind, and TypeScript application and is seeking contributors.

sloptrim is a local prose linter and coding-agent plugin that detects patterns associated with AI-generated writing when an agent saves a document. It extracts prose from supported text and office formats, scores it on a 0–100 scale, and reports suspicious spans for revision; it analyzes writing patterns rather than determining authorship and does not process code. The command-line tool uses Python's standard library, requires no model or network connection, and keeps text on the local machine. It integrates with Claude Code and can be initialized for other agents through an agent contract and editor-rule files.

TrueForge is an open-source, vendor-neutral agent harness developed by TrueFoundry. It runs the agent execution loop, including model calls, MCP tool use, skills, sandboxed code and file execution, approvals, context management, and session state. Agents can use configured model providers, remote MCP servers, git-backed SKILL.md instruction packs, and on-demand sandboxes; context features include subagents, deferred tool loading, Code Mode, large-result offloading, and compaction. TrueForge exposes agents through a bundled chat UI, an HTTP API with a TypeScript SDK, and an embeddable UI SDK. It supports local mode with SQLite and hosted deployments using Postgres and Redis through Docker Compose or Helm; the repository warns that local mode is intended for personal use on localhost rather than production or internet-facing deployment.

unlazy is an agent skill for enforcing completion discipline in substantial AI-agent tasks. It requires an acceptance ledger to be written first, with runnable CHECK commands and EXPECT conditions; its gate checker reviews commands, executes approved gates, records evidence, and can reverify completed work. The skill also supports a Depth Tree method that splits tasks into fresh-context subtasks with integration checks, giving each leaf the full task time budget. It can be installed for supported agents with the skills CLI or manually for Claude Code and Codex CLI; its checker and optional hook require Node.js 16 or newer and no third-party runtime packages.

Vibe ASO is a Claude Code skill that takes an iOS app from ready-to-submit status through App Store optimization and localization. It performs popularity- and difficulty-driven, intent-matched keyword research; generates App Store metadata in up to 50 locales; renders localized screenshots with Playwright, per-script Noto fonts, right-to-left handling, and automatic text fitting; applies territory-specific pricing through the App Store Connect API with read-back verification; and localizes in-app strings. Its checks detect translation problems such as format-specifier drift, wrong-sense domain terms, script contamination, and register drift. It sets and verifies App Store Connect fields where the API permits, producing an explicit manual-steps list for the remaining submission tasks. The skill has no separate SaaS signup: it uses the user's App Store Connect API key, Claude subagents by default, or a DeepSeek/OpenAI-compatible translation API, and can use ASO keyword data from an external tool such as Astro's MCP.

VibePulse is an always-on ESP32-S3 AMOLED panel and local-first host service for monitoring Claude Code and Codex usage, live agent activity, and prompts that need user input. The panel displays quota information and agent status, raises a full-screen NEEDS YOU alert when an agent is waiting, and can send supported decisions back through a tap on the display. A pure-standard-library Python service runs on a macOS or Windows computer and communicates with the panel over the local network. The project supports optional, default-off relays: a numbers-only relay for quota data, an encrypted interaction relay for supported NEEDS YOU decisions, and an independent status relay for Claude/Codex activity. The repository describes the core panel loop, Windows host lifecycle, and physical prompt-answer flow as verified, with cloud features disabled by default.

Viscose is a portfolio carousel rendered as a single full-screen WebGL fragment shader rather than a grid of DOM elements. Project cards travel around a mostly off-screen ring that can be turned by scrolling, dragging, swiping, or clicking; momentum carries the ring and it snaps the nearest card to the front. Signed distance fields make neighboring cards melt together as they approach and form thin, sagging threads as they separate, while pointer interaction softens the field, shifts nearby cards, and creates connecting effects. The same shader pass renders the threads, the refracting glass-like edges at the top and bottom of the viewport, and the cursor tag. On touch devices, swiping turns the ring and press-and-hold triggers the hover effects. The project includes a development-only lil-gui panel with roughly 136 live parameters and is built with Next.js, React, Three.js, GSAP, and Tailwind CSS; it requires Node.js 20 or newer to run locally.

Zeron is a local control layer for coding agents, including Claude Code, Codex, Cursor, Grok, Hermes, and Pi. Each device runs a small engine that stores agent sessions locally by default, without requiring an account or network connection. Optional synchronization lets users start an agent on one device and follow or control it from another; an always-on machine such as a VPS can keep agents running after a laptop is closed. The Linux installer starts a daemon that persists across reboots, while macOS users can install the desktop release or build from source. Zeron is licensed under the MIT License. The videos refer to the project as Comet.