Top Open-Source GitHub Projects : fmt, Tile Language, SkillSpector, QDuo, seiso & DroidDeck Transcript, AI Summary & Key Points

Agent Office is a self-hosted 3D workspace for coordinating coding agents across GitHub repositories. It creates a floor for each project, places Claude Code, Codex, OpenCode and other supported agent workers at desks, streams their terminals, and lets users assign issues, queue tasks, provide Git worktrees, inspect changes and open pull requests. Workers can also list, hire, message and dismiss other workers through the agent-office MCP server or command-line interface. The office includes shared chat, voice, screen sharing and a whiteboard, plus a 2D /lite interface for phones and slower computers. It runs locally or can be deployed to AWS, Azure, Railway, Fly.io, Dokploy or an Ubuntu/Debian server, with multi-user accounts and per-user Claude and GitHub sign-ins. The repository describes it as work in progress, subject to breaking changes, and distributes it under the MIT license.

Agent Substrate is an open-source runtime and control plane for running agent-like workloads at high density. It manages the lifecycle of sandboxed actors, including creation, destruction, suspension, and resumption, and assigns them to a smaller pool of worker pods while routing traffic to the active workers. The system uses Kubernetes for infrastructure provisioning and pod lifecycle management, with support for sandbox technologies including microVMs and gVisor. It preserves actors' volatile memory and filesystem state through full-state snapshots, enabling sub-second suspend and resume operations and multiplexing many stateful actors across shared infrastructure. It manages standard OCI containers at the kernel level and is designed to support different agent frameworks and harnesses, including ADK-compatible actors, LangChain agents, coding environments, and MCP servers. Agent Substrate is intended for running agents at scale rather than building them, and its workloads do not have to be literal AI agents. The repository notes that the project is not an officially supported Google product.

A Claude Code skill for creating motion-design videos in code using HTML, Playwright, ffmpeg, Python, and Chromium. It computes each frame from a deterministic seek(t) function, renders frames with Playwright, blends subframes for motion blur, and encodes the result with ffmpeg. Its motion system includes closed-form springs, log-space camera keyframes, shared-element handoffs, masked word-by-word text, color floods, and ports of selected 21st.dev components. The workflow moves from inputs and a beat map through still-frame approval, full rendering, pop detection, audio generation, and final muxing. Audio processing finds musical drops by band energy, aligns the song to a key visual frame, places sound effects at measured peaks, and applies two-pass loudness normalization. The repository also includes a remake mode for frame-locked recreations with shot analysis, side-by-side verification, and royalty-free replacement audio. The code is MIT-licensed; downloaded third-party media remains under its respective license.

Coucou is an open-source desktop companion for Claude Code and other coding-agent sessions, appearing in the macOS notch or as a top-of-screen island on Windows. It displays session events such as files read, edits, and commands run; surfaces Claude Code permission requests for Allow/Deny approval; opens the relevant terminal on macOS; provides built-in Claude chat; and accepts dropped files as context. It can also attach a macOS window as Claude context and display optional integration pills for services including GitHub, Vercel, Notion, and others. On macOS, a hook script forwards Claude Code events over a Unix socket and waits for approval responses, while a borderless NSPanel and a SwiftUI Canvas render the animated Mochi character. The native Swift 6, SwiftUI, and AppKit application has no third-party dependencies. On Windows, a Tauri 2 application uses Rust, TypeScript, a transparent always-on-top window, Canvas 2D rendering, and a named pipe for hook events. The project states that it collects no telemetry, requires no account, and stores credentials in the macOS Keychain or Windows Credential Manager. Coucou is distributed from the project's releases and can also be built from source. The code is licensed under MIT; the Mochi name, character, icon, sounds, and media are separately reserved by the author. The Windows installer is temporarily unavailable according to the README because its unsigned build was flagged by Microsoft Defender.

DroidDeck is an Android application that brings a SteamOS-like gaming interface to supported Adreno handhelds. It runs Valve's Steam client in Big Picture mode and uses ARM64 Proton to run Windows games; an optional Linux desktop provides Flatpak applications and games from Flathub, along with emulators. It requires Android 9 or newer and a supported Adreno device, does not require root, and is distributed as an APK with a separate Linux runtime. Compatibility and performance vary by device; desktop compositing uses software rendering, and Firefox sandboxing is reduced under proot. The project is licensed under GPL-3.0 and is not affiliated with Valve.

ECC is an MIT-licensed open-source agent-harness performance system maintained by affaan-m. It packages reusable skills, specialized agents, project rules, commands, hooks, memory, and security tooling for Claude Code as its primary target, with supported or limited adapters for Codex, Cursor, OpenCode, Gemini, Zed, GitHub Copilot, and other coding harnesses. Its core workflow turns plan, test, implement, review, verify, remember, and improve into reusable agent workflows. Skills are loaded for tasks such as test-driven development, research, security review, end-to-end testing, documentation, and refactoring; agents isolate planning, implementation, and review; rules provide always-loaded project or language standards; and hooks run event-triggered checks and session automation outside the model context. The optional Memory Vault stores inspectable Markdown handoffs and session context, while AgentShield scans agent files, hooks, MCP configurations, permissions, prompts, and secrets for security risks. ECC can be installed from its repository or through the ecc@ecc Claude Code plugin and ecc-universal package. The repository also provides selective installers, native or project-local integrations for several harnesses, a desktop dashboard, and the optional ecc-agentshield security-auditing package. Feature parity varies by harness: GitHub Copilot receives instructions and reusable prompts but not ECC hooks or agent delegation, while Codex has a native marketplace plugin with a narrower hook model.

FreeLLMAPI is an open-source proxy that aggregates free tiers from multiple LLM providers behind a single OpenAI-compatible `/v1` endpoint. It can also connect custom OpenAI-compatible chat, embedding, image, and audio endpoints, including local servers and remote gateways. Its router selects an available model, fails over when a provider is rate-limited, stores provider keys encrypted, and tracks per-key usage against free-tier limits. The router refreshes its model catalog from a signed feed; free installations receive periodic snapshots, while the hosted premium offering provides the live catalog. The project is intended for personal experimentation and is available as a self-hosted install with a hosted service at freellmapi.co.

HyperFrames is an open-source framework for turning HTML, CSS, media, and seekable animations into deterministic MP4 videos. It can be used locally through a CLI, by AI coding agents through skills, or as a rendering core for hosted authoring workflows. Compositions are defined as HTML with data attributes for timing, tracks, media, and sub-compositions. The renderer seeks each frame in headless Chrome, uses animation adapters such as GSAP, CSS, Lottie, Three.js, Anime.js, or WAAPI, and encodes the result with FFmpeg; its toolchain includes preview, linting, inspection, local rendering, browser-based editing, reusable catalog components, and AWS Lambda rendering. The project requires Node.js 22 or newer and FFmpeg, does not require React or a build step, and is licensed under Apache 2.0. Its agent skills support workflows including product videos, explainers, pull-request videos, captions, motion graphics, slideshows, music-driven videos, and general compositions.

An agent skill for Claude that guides logo projects from discovery briefs through concept development, SVG construction, testing, refinement, and delivery of brand assets. Its workflow researches category conventions in a classified library of more than 1,400 reference SVG logos, develops 8–12 concepts, builds selected concepts in SVG, and pauses at a checkpoint for direction selection before producing a full kit. Dependency-free Python tools audit SVG structure and craft rules, generate concept and test sheets, check marks at sizes down to 16 px in one-colour and reversed treatments, compare them in contextual and competitor-shelf tests, render PNGs, create presentation boards, and export favicon, app-icon, web-manifest, lockup, and other variants. It can also produce icon sets and brand-guideline materials. The skill follows the Agent Skills format, supports Claude Code, Claude.ai, Claude Desktop, and compatible agents, requires Python 3.8+ for its tools, and is distributed under the MIT license; the reference logo files are third-party trademarks excluded from that license.

PaperMono Shopping List is an open-source household shopping-list application consisting of firmware for the M5Stack PaperMono e-paper device, a FastAPI and SQLite server, and a phone web UI. The device displays items grouped by aisle, supports touch-based ticking and on-device entry, and works offline by queuing edits on flash. Wi-Fi is enabled only for scheduled synchronization, after edits, or when a stale list is accessed; the phone connects to the server over HTTP. The server maintains item and aisle data and can optionally send new item names to the Claude Code CLI for aisle classification, while manual aisle corrections are retained. The repository is licensed under the GNU GPL v3.0 and describes the project as a personal setup rather than a commercial product; it has been tested only with the PaperMono C153 and the server has no built-in authentication.

QDuo is a macOS menu-bar utility that displays configurable actions at the cursor when text is selected in native apps, browsers, Electron apps, or PDFs. Actions can translate, polish, explain, summarize, search, speak, transform, run shortcuts or shell scripts, or send prompts to DeepSeek, OpenAI, Doubao, Qwen, Ollama, and other OpenAI-compatible endpoints. It can also extract text from a screen-region screenshot. Results can replace the selection, be inserted after it, or be copied to the clipboard. The app supports capsule or ring-shaped popups, restores the clipboard when it must use clipboard access, stores API keys in the macOS Keychain, and uses a JSON configuration file at ~/.config/qduo/config.json. It is distributed as a signed and notarized macOS application through Homebrew or GitHub Releases, requires Accessibility permission, optionally requires Screen Recording for screenshot text, and is licensed under GPL v3.0.

ReelMimic is a local 2D animation tool that uses Claude Code or Codex to create a new video from a reference video and a brief. It analyzes the reference's shot count, pacing, BPM, transitions, framing, colors, and camera movements, then produces a storyboard, character and asset plan, and style frames for approval. After approval, multiple AI agents build different parts in parallel, while separate agents review completed shots and request fixes with before-and-after screenshots. The tool supports seven drawing engines: vector/motion graphics, hand-painted watercolor, crayon picture book, pixel art, paper cut-out stop-motion, whiteboard doodle, and anime cel; new styles are defined in Markdown. It runs on the user's computer, uses the user's Claude Code or Codex account, and does not reuse the reference video's footage, characters, logos, or assets. The repository is MIT-licensed and has been tested primarily on Windows, with macOS and Linux support described as less tested.

reladraw is a text language and command-line tool for creating diagrams with relative positioning. It sits between automatic layout systems such as Mermaid, Graphviz, and D2 and manually positioned editors such as draw.io: users declare nodes, relationships, labels, and relative placement without specifying absolute coordinates. The parser and layout engine resolve the declarations into a diagram, while the tool reports issues such as overlaps, crossed edges, and text overflow without requiring a rendered preview. Its TypeScript CLI converts `.reladraw` files to SVG, and the repository also provides a browser version and an optional agent skill. The project is early-stage, its syntax is subject to change, and the code is licensed under Apache-2.0.

seiso is a Markdown convention and linter for project documentation written by AI and read by people and coding agents. Its convention assigns each document a kind, such as how-to, reference, or ADR; keeps each fact in one canonical location; requires pointers to named files or symbols; separates different concerns; and avoids rapidly changing values in long-lived pages. The reference implementation checks documents against the convention and reports where each violation occurs and how to fix it, including the decision needed when a judgment call is involved. Stable rules run by default, while experimental preview rules require an explicit option; the CLI also provides commands for initialization, rule explanations, parsing, and checks. It can be installed through Cargo, PyPI, npm, or pipx, and integrates with Claude Code hooks, pre-commit, and CI. The project is licensed under MIT.

SkillSpector is an open-source security scanner for AI agent skills, developed by NVIDIA. It scans Git repositories, URLs, ZIP archives, directories, and individual skill files before installation, looking for prompt injection, data exfiltration, privilege escalation, supply-chain issues, dangerous code, MCP permission problems, tool poisoning, and other malicious patterns. Its two-stage pipeline combines static analysis—including regex, Python AST, taint tracking, YARA signatures, and OSV.dev dependency lookups—with optional LLM-based semantic evaluation. It produces terminal, JSON, Markdown, or SARIF reports containing findings, severity labels, a 0–100 risk score, and installation recommendations. Baselines can suppress accepted findings, and the CLI can be used in CI or install gates through exit codes and machine-readable output. SkillSpector can also run as an MCP server callable by compatible agents. It does not execute scanned skills or sandbox them; optional LLM analysis sends eligible file contents to the configured provider, while dependency checks query OSV.dev. The project is distributed under the Apache License 2.0 and is part of NVIDIA's Verified Skills pipeline.

TileLang is a Pythonic domain-specific language and compiler for developing high-performance GPU, CPU, and NPU kernels such as GEMM, quantized matrix multiplication, FlashAttention, and linear attention. Built on TVM, it uses tiled programming primitives for memory transfers, pipelining, matrix operations, parallel elementwise work, layouts, and target-specific code generation; its JIT specializes kernels for input shapes and compile-time arguments on first use. The project supports CUDA, ROCm/HIP, Apple Metal, Huawei Ascend, and experimental LLVM, CuTe DSL, and WebGPU backends, with additional ecosystem adapters for other accelerator platforms. It is distributed as a Python package through PyPI, with source builds and nightly wheels available.

UniMate is an AI research system for text-to-motion generation across diverse 3D skeletons, developed by researchers at Princeton, UC Berkeley, MIT, and NTU. Given a rigged asset and a text prompt, it generates articulated motion without per-skeleton retraining or test-time optimization. The system is trained on the UniML3D dataset, which canonicalizes text-paired motion for bipedal, quadrupedal, avian, marine, insectoid, serpentine, and articulated rigid-object skeletons. Its training pipeline uses flow matching: a network predicts the velocity of a linear interpolation between noise and motion data, with masked L2, geodesic rotation, and velocity-smoothness losses. The model supports graph-based spatial and temporal attention or full attention over joint-time tokens, with text conditioning through adaptive layer normalization or cross-attention. At inference, classifier-free guidance generates motion features and can render skeleton animations or export animated GLB and FBX files. The same trained model supports text-to-motion generation, motion in-betweening, text-guided joint-preserving editing, and expansion of sequences from multiple prompts. The repository includes dataset processing, training, inference, and preprocessing code, and is released under the MIT License. The included datasets and source assets remain subject to their original licenses.

Universal Modder is an open-source toolkit by rehan-remade for AI coding agents that create offline, single-player mods for PC games. Its skills and tools identify a game's engine and existing modding route, inspect code and files, back up saves, build a working mod slice, generate art, 3D assets and audio through the fal MCP server, test the result in the running game, package it, and record findings for later agents. The repository supports Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode and other agents that read AGENTS.md, and includes the Python `um` CLI for game scanning, asset generation and processing, Windows automation, save management, video capture, publishing checks and a shared knowledge base of game-specific field notes. Its rules restrict use to games owned by the user and offline or single-player work; they prohibit online cheats, anti-cheat or DRM bypasses, and distributing game files or decompiled code. The repository is MIT licensed and requires Python 3.10+ and ffmpeg; Blender is used for 3D-to-sprite rendering.

Zulu is a native Zulip client for iOS and macOS. It treats topics as first-class: topic-based channels appear as lists of threads, while channels without topics use a plain chat view. Users can organize channels into personal folders, promote topics alongside channels, and rename or hide items locally without changing the Zulip server. The client supports reading and sending messages in channels, topics, and direct messages; replies, reactions, file and image attachments, camera captures, and polls. The repository includes shared SwiftUI code, Zulip API and synchronization packages, a macOS app with sidebar and split-view navigation, and a separate Go notification service that builds to a container but is not deployed. It is described as early but usable and supports sign-in to any Zulip server with password or SSO authentication.