tierlist.tech
中EN

09.30 Today's Picks

377 new projects found today, 218 rated.

Created 2026-09-28 – 2026-09-30Fetched 600Filtered 50Rated 41

S · Look today

S

mcp-extensions

openaiTypeScript

OpenAI's officially open-sourced ChatGPT plugin development kit, bringing MCP capabilities into ChatGPT as native features—a platform-level entry point. With 208 stars in a day and OpenAI's backing, it will almost certainly become a de facto standard dependency for the ChatGPT plugin ecosystem, with clear monetization (plugin marketplace, API billing). The risk is ecosystem policy changes and cross-vendor standard fragmentation.

Build plugins that feel like native, first-class features of ChatGPT.

#MCP#SDKs#Plugins#AI Agents

★208

A · Worth watching

A

AIHOT

KKKKhazixTypeScript

Solves the repetitive work of aggregating RSS and social sources into an industry daily-news site, with a framework design where changing sources changes the site. 3,000+ stars in a day validates demand for automated content sites. The business path is an intelligence briefing SaaS for vertical media and enterprises, but LLM-written content is highly homogenized, and the biggest risk is general AI summarizers and RSS aggregators siphoning the value.

A website framework that finds trending topics on its own and writes its own daily digest. Swap in your own sources and curation criteria, and it becomes your industry trending site.

#Content Writing#Automation#AI

★3087

A

jeeves

PostHogPython

PostHog's official account builds inference enhancement for Jev-style decision models, with 267 stars in a day and backing from a well-known vendor—the most endorsed of this batch's 'Jev' signals. If Jev becomes a lightweight decision paradigm, the inference layer could be absorbed by agent frameworks as a standard component. The risk is an overheated Jev ecosystem with unvalidated real demand and no independent user data.

Jeeves – Reasoning improves Jev-like decision models

#AI#ML Research#AI Agents

★267

A

tusk

alpcanaydinRust

A macOS-native, keyboard-first database client supporting 20 databases plus an AI assistant, directly targeting the paid market of TablePlus and DBeaver where demand is rigid and willingness to pay is clear. Rust-native performance and open-source freedom are the differentiators, but the database-client space is mature with high switching costs, so it must open a gap through AI-assisted querying.

A fast, native, keyboard-driven database client for macOS. 20 databases, an AI assistant that shows its work, free and open source.

#Databases#macOS#AI#Productivity

★68

A

parquet-xray

cfahlgren1TypeScript

A Parquet file inspector that lays the bytes out for you, aimed at data engineers debugging encoding, compression and page layout. The scope is small and specific, filling the gap where parquet-tools gives only metadata and no byte view. The library is still small, but the direction is clear.

See how a Parquet file is laid out, byte by byte

#Data Analysis#Debugging#Web Frontend#parquet

★28

A

homedroid

sudiptacmdKotlin

Turns an idle Android phone into a home server without root: SSH, web hosting, Jellyfin hardware transcoding, Immich, qBittorrent, Home Assistant and a dashboard, with a checklist concrete enough to follow. Very practical for anyone wanting a free low-power NAS. The downside is that no one guarantees long-term stability or cooling.

Turn an old Android phone into a home server: SSH, web hosting, Jellyfin with hardware transcoding, Immich, qBittorrent, Home Assistant and a web dashboard. No root.

#Android#Self-Hosted#DevOps#jellyfin

★19

A

elm-compiler

SpotttRust

A Rust rewrite of the Elm 0.19.1 compiler, published as a drop-in replacement on npm, with a narrow scope and restrained claims. A real option for people still maintaining Elm projects who find the official compiler slow; the risk is a stagnant ecosystem with long-term maintenance by one person.

Fast, unofficial Elm 0.19.1 compiler written in Rust — drop-in replacement for elm (npm: @spottt/elm-compiler)

#Build#Performance#CLI

★18

A

claude-sap-kit

silvius1996ABAP

Connects Claude Code to SAP: running ABAP reports, previewing NAST print output as PDF, and editing SmartForms as XML. A real time-saver for ABAP developers who tweak reports daily, and a far more specific entry point than general coding assistants. The limitation is dependence on its own ARC-1 extension, with thin docs and ecosystem.

Claude Code plugin + ARC-1 extension + ABAP kit: run SAP reports, preview NAST print output as PDF, edit SmartForms as XML

#Plugins#AI Agents#Automation#sap

★12

A

synpath

Synpath-aiPython

Unifies Kalshi and Polymarket market data and order placement into one Python SDK, with tick-by-tick order book history. It's convenient infrastructure for prediction-market strategists; the cost is heavy dependence on both exchanges' APIs, with stability yet to be verified.

One open-source API for prediction markets. Trade Kalshi, Polymarket and Polymarket US with one Python SDK: unified market data, smart order routing across venues, and advanced order types. Plus tick-…

#Fintech#APIs#SDKs#Crypto

★16

A

ThermalVortex-Mod

shamumumuLanguage unknown

An unofficial character mod for Slay the Spire 2 with 119 cards, a monster field, fusion summons and a meta reward pool, plus a directly installable package, making it an immediately usable sample for Chinese players and mod authors. The limitation is tight coupling to a single game version, worthless outside that circle.

Yugi Muto | Unofficial character mod for Slay the Spire 2. 119 character-related cards, monster field, Fusion Summon, and meta reward pool building, with a directly installable mod pack.

#Games#Game Dev#slay-the-spire

★12

A

vita-flash-emulator

ibrahim-doganRust

Brings Ruffle to the PS Vita; one VPK install lets you pick games and play Flash, with touchscreen, button mapping and a pause menu. The platform is niche but the need is real, and it honestly notes it is based on Ruffle. The limitation is that Flash content itself is dying, so the audience is limited to Vita homebrew players.

A Flash player for PS Vita, built on Ruffle. One VPK: launch it, pick a game from your library, play — with touch, button mapping and a pause menu.

#Games#Embedded#flash#ruffle

★16

That's today. The other tiers and the boards follow.

BHas a spark31
  • B iCode Positioned as an offline-first AI development platform plus agent/workflow toolbox, TUI-friendly with self-held data, hitting the real need for localized agent development. 238 stars in two days is a good start. Monetization can be open-source core plus enterprise private deployment, but the track is extremely crowded (Cursor, OpenClaw, various CLI agents), and insufficient differentiation is the biggest risk. #AI Agents#Self-Hosted#CLI

    A lightweight, extensible, fully offline development platform and agent/workflow toolkit for the AI era, with an intuitive TUI, easy deployment, and complete control over your data, agents, and workfl…

    openJiuwen-aiPython★238

  • B oxbox A tiny JS/TS/JSX parse-and-transform library from unjs built on oxc, filling a gap for lightweight AST tools in the JS ecosystem as a Babel alternative an order of magnitude faster. unjs has strong trust in the Node toolchain and is easily absorbed as a standard dependency; the risk is the fast-evolving oxc/sWC ecosystem requiring continuous maintenance. #Build#Code Quality#Web Frontend#oxc

    🐂 Tiny, portable JS library for parsing and transforming JS/TS/JSX, powered by oxc

    unjsTypeScript★39

  • B esp-sdr Turns a cheap chip into an SDR using ESP32's undocumented raw I/Q capture, a genuine hardware-level hack with high novelty. It could spawn low-cost radio monitoring and teaching networks, but the risk is undocumented reverse-engineered features being killed by firmware updates anytime, low duty cycle limiting practical use, and commercial use requiring dedicated SDR chips. #Hardware#Embedded#Network Security#Reverse Engineering

    ESP-SDR uses the undocumented raw I/Q capture functionality of ESP32 family chips to turn them into (low-duty-cycle) Software Defined Radios

    ESPARGOSC★33

  • B heapspace Uses Three.js and ClrMD to turn .NET memory dumps into rotatable 3D visualizations, hitting the pain that heap structure is hard to grasp intuitively in .NET performance troubleshooting. The .NET ecosystem lacks such visualization tools, making this scarce supply, with a path to becoming a paid diagnostic tool like JetBrains or Razorpop. The risk is a narrow audience and a .NET community smaller than JVM's. #Debugging#3D#Data Analysis#dotnet

    Heapscape: a local Three.js and ClrMD explorer for .NET memory dumps

    kkokosaJavaScript★74

  • B RuiC-motion-reel Generates 15-second animated films from code, with vector supersampling, point cloud and color-separation rendering, outputting 1920×1080/30fps plus a finished-film directory. Suited to people making data-visualization shorts without opening AE; render quality and the actual effect of the "15 styles" need the finished films to judge. #Video#Animation#AI Agents#Agent Skills

    Generate 15-second motion graphics films with code · Agent Skill. Delivers 1920×1080 / 30fps. Self-built rendering engine: 2× supersampled vectors/text + true 3D point clouds + screen-print four-color…

    HRuiCccPython★16

  • B nuyoah-outfit-asset-generator Generates a full asset board with separated parts from clothing reference images, directly serving game and virtual fashion production pipelines, where part-splitting consistency is a real industry pain. As an MIT Skill it can embed in agent workflows and could become a standard skill for digital asset generation. The risk is unstable generation quality and part-splitting precision, with similar 3D clothing generation approaches evolving fast. #Agent Skills#Image Generation#Video

    MIT Skill that generates a full outfit and a separated-parts asset board from a clothing reference image

    nuyoah-ai-worksLanguage unknown★40

  • B local-enough Estimates whether a background AI task can run on your own hardware: accuracy, p50/p95 latency, cost per thousand runs, and break-even volume, plus a router that picks the cheapest model meeting a quality bar. It's practical for small teams making localization decisions; the catch is needing labeled data first. #LLMs#Benchmarks#Self-Hosted#Performance

    Measure whether a back-office AI task can run on your own hardware: accuracy vs gold, p50/p95 latency, $/1k tasks, amortised hardware cost and break-even volume, plus an OpenAI-compatible router that …

    B0ykoPython★8

  • B jevwright End-to-end tests written as user operation steps, where AI only locates controls the first time and then runs via recorded playback, with assertions still decided by code. Attractive to QA teams tired of fragile selectors; but there's no data on locator accuracy or playback stability, and strong rivals like Playwright are ahead. #Testing#AI Agents#Automation#Web Frontend

    Browser tests for business flows, written as the steps a user takes. AI finds the controls once, recordings replay for free, and your code decides the verdict.

    Ice-HazymoonTypeScript★9

  • B TIRx-harness A compiler harness from mlc-ai that aims to let agents automatically write GPU kernels and verify them on the fly. From a trusted team and genuinely novel in closing the loop on compiler feedback, but it is one day old with almost no docs, so for now you can only watch, not use it. #GPU#ML Research#AI Agents

    An Open Compiler Harness for Agentic GPU Programming

    mlc-aiRust★18

  • B robolings A rustlings-style robotics exercise set using only NumPy, covering kinematics, dexterous hands, diffusion policies, and flow matching, suited to people who want hands-on practice rather than reading papers. Its restrained scope and lack of heavy dependencies are strengths, but the exercise set is just starting with no automated grading loop, so completeness remains to be seen. #Robotics#Learning#ML Research

    Small exercises for robot learning, rustlings style: kinematics, dexterous hands, control, diffusion policy, flow matching. NumPy only.

    12-Twelve-12Python★10

  • B XelImageFormats---50-image-formats-for-TPicture-webp-jp2-heif-avif-heic-psd...- Adds decoding support for webp, heif, avif, psd and 50+ other formats to Delphi's TPicture, with a concrete need and clear boundaries. Such libraries are scarce in the Delphi ecosystem and it works once installed; the cost is that it only matters to Pascal desktop developers. #Images#SDKs#Plugins

    XelImageFormats - 50+ image formats for TPicture (webp, jp2, heif, avif, heic, psd...)

    XelitanPascal★8

  • B pixeltop A real-time system monitor for the Omarchy desktop, emphasizing theme-matched colors and inspired by btop. A single Go binary, install-and-go, with clearly stated boundaries; the only selling point is theme adaptation, since such monitors are plentiful. #CLI#Terminal#Monitoring#Linux

    A live system monitor for Omarchy that puts your theme first. Inspired by btop.

    bjarneoGo★19

  • B Nostalgex Wraps Plex, Jellyfin and Emby libraries into a retro cable TV schedule, with a native tvOS app plus a web tuner, sure to amuse self-hosting fans. The idea is likable and the boundaries are clear, but it is only an early version, and library compatibility and actual experience are unverified. #Streaming#Self-Hosted#UI Design#plex

    Retro cable TV guide for your Plex, Jellyfin or Emby library. Native Apple TV app + web tuner.

    chadMuellerSwift★16

  • B Scrubbers A set of twelve feel sliders for SwiftUI, callable with one view plus one enum, with restrained naming and scope. It is not a new problem, but it breaks down drag feel cleanly, so iOS interaction people can copy it directly; the limitation is that it is more a visual experiment than a general control library. #iOS#UI Design#SDKs#swiftui

    Twelve sliders you'll want to touch, for SwiftUI. One view, one enum: Scrubber(.jelly, value: $value). Liquid Glass, jelly, elastic, fluid, squiggle, thermostat, swing, mood and more.

    haplollcSwift★14

  • B chatgpt-receipts A Chrome extension that labels each ChatGPT reply with the model that actually answered, and warns when it's silently switched. A small, precise pain point for people who depend on a specific model; the limits are that it only covers the web UI and relies on page structure, so it may break at any time. #Browser Extensions#LLMs#Privacy

    Chrome extension that shows which model actually answered each ChatGPT reply, and flags it when you get rerouted.

    Lex-auJavaScript★10

  • B OpenRayneo A Bluetooth bridge for RayNeo iO glasses on macOS, providing local HTTP notifications and teleprompter control. It targets a specific gap in specific hardware, with use and limits clearly stated; the audience is only Mac users of these glasses, but for them it's nearly the only choice. #macOS#Hardware#APIs

    Unofficial macOS Bluetooth bridge for RayNeo iO glasses: local HTTP notifications and teleprompter control.

    Tobi1chiSwift★9

  • B limen A local proxy in front of coding agents that tracks what they resend and compiles scaffolding out of prompts, non-intervening by default. The idea goes a step deeper than typical token panels; for now it's only measurable, so you'll have to test how much context it saves. #AI Agents#LLMs#Observability#CLI

    Local, opt-in proxy for coding agents: measures what your agent re-sends, compiles the scaffolding out of the prompt, and enforces nothing until you say so.

    aashish254Python★10

  • B china-market-data Wraps A-share quotes, candlesticks, fund flows and announcements into a CLI skill for agents, a pragmatic direction: letting Claude Code query data directly. Long-term usability depends on the data source, which is not explained; similar needs are currently mostly handled by libraries like akshare. #Fintech#Data Analysis#CLI#Agent Skills

    China market data / China stock market data Skill: A-share quotes, K-lines, tick data, research reports, capital flows, announcements, ETF/futures/macro. A Python CLI for Claude Code and AI Agents. Ch…

    clawyi-comPython★17

  • B xiaomi-LYWSD03MMC-to-solar-meter Turns an idle Xiaomi temperature and humidity sensor into a real-time photovoltaic power meter, showing generation power and self-use ratio in Home Assistant. Extremely niche but extremely specific, exactly the kind of hardware hack you can follow after ten minutes; the prerequisite is having the same sensor and solar setup. #IoT#Self-Hosted#Hardware#home-assistant

    Turn a Xiaomi LYWSD03MMC humidity monitor into a live solar meter for Home Assistant – PV output in kW and self-consumption %

    jozefpisPython★11

  • B steno A local Whisper dictation tool for macOS that records while you hold a hotkey, also handling meeting notes, ready to use once installed. The need is real but the field is crowded, with VoiceInk, MacWhisper and Superwhisper all doing the same thing, and no differentiator is visible here. #Speech#macOS#Productivity

    Free, open-source dictation and meeting notes for macOS. Hold a key, speak, and your words appear at the cursor – transcribed locally with OpenAI Whisper.

    ThesimpleexSwift★8

  • B iigs-doom Running Doom on the Apple IIgs, requiring a ZipGS or TransWarp GS accelerator—a hardcore retro port. Worth a look for those into the 65816 platform and extreme ports; practicality is near zero and few can run it on real hardware. #Game Porting#Games#Hardware

    Can it play Doom? Doom on the Apple IIgs with a ZipGS or TransWarp GS accelerator.

    WebifiAssembly★8

  • B pi_plays_pokemon Lets a coding agent play Pokemon from pixels and keypresses, essentially an end-to-end agent sandbox. The gameplay is not original, since Claude and Gemini have played it, but the code is readable and can be turned into your own agent benchmark, and with the author's reputation it is worth ten minutes. #AI Agents#LLMs#Games#Benchmarks#pokemon

    pi (coding agent) plays Pokemon from buttons to pixels

    geohotPython★18

  • B noterepo A local-first macOS journal app written by Flavio Copes, who has a long writing reputation, so the code style is probably fine. But the local notes space is already packed tight; unless it clearly differs in launch speed and pure keyboard flow, there is no reason to migrate. #macOS#Productivity#Documents

    A quiet, local-first daily notes app for macOS

    flaviocopesSwift★22

  • B Sims3-ApexRadiance A lighting and performance mod for The Sims 3, with a description specific enough to mention nighttime light placement, water reflections and multi-core texture compression, showing real work on an old engine. For people still playing the third game it is a solid improvement, but it piles on many features, so stability must be tested yourself. #Games#Performance#3D

    Lighting, visuals and performance mod for The Sims 3. At night, lamps really light the ground, lots, walls, roofs and snow; ponds glow and mirror their shore. Fewer stutters: faster file lookups, lot …

    loinyxC++★8

  • B vocab-flashcards A single-file HTML flashcard app: parses PDF/PPTX/DOCX locally to extract words, has a built-in offline dictionary, and looks up words via Baidu and Youdao. It's genuinely ready to use for vocabulary learners, but similar tools are plentiful; its only edge is direct access in China and single-file distribution. #Learning#Documents#Productivity

    Single-file English vocabulary flashcard tool: locally parses PDF/PPTX/DOCX to extract words, with a built-in offline vocabulary and online lookups via Baidu Translate and Youdao Dictionary through do…

    Jeng-linLiHTML★20

  • B adu-motion-video An agent skill that assembles narration, subtitles and audio into editable HTML/JS scenes and exports local MP4, with export validation—more substantial than pure prompt collections. Still, it's in the crowded short-video auto-generation lane, and output quality is unclear. #Agent Skills#Video#Video Editing#AI

    AduNext motion-video skill: editable HTML/JS scenes, talking-head layouts, captions, audio and verified local MP4 export.

    adunextPython★17

  • B steam-frame-nixos Gets a Valve Steam Frame VR session running on NixOS, an experiment explicitly labeled a spike, honest and technically substantive. It has reference value for the NixOS-and-VR intersection, but coverage is extremely narrow and upstream changes could topple it anytime, making it more notes than tool. #Linux#Hardware#Games

    NixOS with Valve's VR session on the Steam Frame (spike)

    y0usafNix★30

  • B floorplan-3d From the name, a web tool that turns 2D floor plans into 3D views, a likable direction that hit hundreds of stars in a day, showing real demand. But the repo has no description at all: input format, export support and accuracy are unknown, and stars rising faster than docs is all the more reason to wait for it to explain itself. #3D#Web Frontend#UI Design

    wy51aiHTML★478

  • B zhijian-receipt A pure front-end receipt generator that auto-calculates prices, exports images and high-res PNGs, handy for small shops or people making mockups. Single-purpose but honestly described; there are already many similar online receipt tools, and no differentiation is visible. #Web Frontend#Documents#E-commerce

    Paper Space · Receipt Generator | A simple receipt-making tool with automatic calculation, image import, local drafts, and high-definition PNG export.

    PomecroHTML★8

  • B go-link Lets a computer run MAME and pushes video, audio and voice over WebRTC to friends in a browser, so online arcade play needs no client, and the Go plus Pion choice is fitting. But the claim of not needing ROMs needs explanation, and remote co-op already has mature alternatives like Parsec and Fightcade. #Games#Streaming#Self-Hosted#mame

    Play classic arcade games online with friends, straight from the browser. Your computer runs MAME (libretro) and streams video, sound and voice over WebRTC; friends join with a link and a PIN, no ROMs…

    lordbasexTypeScript★10

  • B milk A lightweight X11 desktop and window manager written in Odin, pitched as beginner-friendly. Choosing Odin for a WM is interesting, but the WM space is saturated, and "beautiful, beginner-friendly" are self-assessments with no config examples to back them up. #Linux#UI Design#Operating Systems

    A lightweight, beautiful, and beginner-friendly X11 desktop and window manager.

    mirvoxtmOdin★11

COrdinary77
  • C indecis A small decision model implemented in Go that offers a high-performance local inference option within the Jev ecosystem, complementing projects like Jevstiller. Go lacks a lightweight decision runtime, so it could become a standard-library candidate, but the whole Jev paradigm is unproven in production and solo-developer maintenance continuity is a risk. #ML Research#Performance#Data Analysis

    Small decision models in Go

    BornholmGo★36

  • C PhysicalCoding A self-evolving coding agent for the physical world, targeting high-value automation in robotics and embedded development with a novel concept. But 38 stars in two days shows no community yet, physical-world validation is costly and slow, and commercialization must tie to robot vendors or simulation platforms, leaving high uncertainty. #AI Agents#Robotics#Automation

    Self-Evolving Coding Agents for Physical World

    HexaFutureLanguage unknown★38

  • C reelmimic Generates new videos in a user's preferred style, with an AI crew of Claude Code and Codex planning and reviewing, targeting the marketing need for stylized short-video creation. But style-transfer video generation is crowded with Runway, Kling and others, and output stability and copyright risk are hard flaws, so real output quality is what matters. #Video#AI Agents#Content Writing#Marketing

    Show it a video you love. Get a new video in the same style. An AI crew (Claude Code or Codex) plans, builds and reviews it with you.

    edenfunfJavaScript★37

  • C emdash-build Uses conversational AI to build sites in a Cloudflare Sandbox and connect them to EmDash CMS, an open-source plus edge-deployment variant of the v0/Lovable track. The CMS plus AI site-building tie-in offers a paid conversion path, and the Cloudflare ecosystem helps hosted distribution. But the track is crowded with giants, so an independent product needs open-source and self-hosting to differentiate. #AI Agents#Web Frontend#Self-Hosted

    Chat-based site builder for EmDash CMS — describe a site, watch an AI agent build it in a Cloudflare Sandbox

    emdash-cmsTypeScript★37

  • C OMEGA-0 ω-0 is an implicit predictive world model for concurrent loco-manipulation in humanoid robots, a frontier direction in robot learning with high academic novelty and signs of paper plus open-source code. The business path is procurement or licensing by humanoid robot makers in physical AI, but the landing cycle is long and depends on sim-to-real transfer, limiting near-term momentum. #Robotics#ML Research#AI

    The offical code of ω-0: A Latent Predictive World Action Model for Concurrent Humanoid Loco-Manipulation

    gentlefressPython★35

  • C stelliberty_android A modern Android network client for on-device packet capture and debugging, offering a lightweight Charles alternative on mobile. Debugging demand is rigid and it could become a small, polished paid tool, but Capillary and HttpCanary already occupy the space, so differentiation and certificate-setup friction are adoption barriers. #Networking#Android#Privacy

    A modern network client for android.

    Kindness-KismetKotlin★33

  • C ai-chatroom Lets Claude, ChatGPT, Grok and Gemini converse autonomously in a web group chat, using already-logged-in CLIs to avoid API keys, a novel and shareable gimmick. It could serve as a demo or research template for multi-agent interaction, but it lacks real productivity use cases, has an unclear business path, and depends on CLI login sessions that easily expire. #AI Agents#Social#Multimodal

    A web group chat where Claude, ChatGPT, Grok and Gemini hang out and talk on their own, powered by your logged-in CLIs (no API keys). English / 한국어 / 日本語

    Moris-krJavaScript★37

  • C Tame_3D Multi-agent 3D reasoning plus scene-level conformal uncertainty alignment, emphasizing 'evidence grounding' to reduce 3D generation hallucinations—a real problem (3D generation reliability). It's a research framework with no mature users or benchmark endorsement yet, needing to compete with various 3D generation and evaluation approaches; commercially it looks like an early B2B 3D content QA tool. #3D#ML Research#AI Agents

    Tame3D research framework: evidence-grounded multi-agent 3D reasoning with scene-level conformal uncertainty alignment

    apromisedlandPython★35

  • C wondersnap A 3D model driven by MediaPipe hand tracking and handwritten WebGL2 with 200,000 GPU particles, with solid in-browser interaction technique. It could be reused for AR/3D product showcases and interactive marketing with some B2B value, but embedding cost as a general library is high, making it more a piece than a dependable component. #3D#Web Frontend#GPU

    Gesture-controlled 3D models made of 200k glowing GPU particles. MediaPipe hand tracking + hand-written WebGL2, runs in the browser.

    AkbarSheikh-debugJavaScript★47

  • C local-ai-recipe-kit Quantifying local LLM benchmarks under a unified protocol and submitting them to local-ai-registry, for people comparing hardware. The idea is clear and well-scoped, but it's currently just a submission channel depending on an upstream registry, lacking its own measured samples; similar benchmark scripts already exist. #LLMs#Benchmarks#Self-Hosted#GPU

    Run a big model on your own GPU, measure it with one protocol, and submit the result. Intake for local-ai-registry recipes.

    0xSeroPython★9

  • C 1c-testpilot-agent-kit An AI testing agent skill pack and benchmark for 1C enterprise systems, operating the UI via ROCTUP/1c-testpilot. The problem is well chosen, but the audience is limited to the Russian-language 1C ecosystem and it needs commercial tools to run. #AI Agents#Agent Skills#Testing

    A skill, tester agent, and benchmark for AI agents that test 1C via ROCTUP/1c-testpilot

    afk-yarHTML★14

  • C omaframe A small tool that adds screenshots and screen recording to Omarchy and Hyprland, with a clearly limited environment and a single job, installable today by Hyprland users. The problem is that grim, slurp and hyprshot already cover similar needs, so it must prove it is genuinely easier for recording and interaction, or it is just another wrapper. #Linux#Video#Productivity#hyprland#omarchy

    Screenshots and screen recordings for Omarchy and Hyprland

    btsouthC++★11

  • C pilulier A one-time prompt fragment for AI companions, burned after use, built as a zero-dependency Web Component, with some cleverness in the mechanism. It is essentially still a collection of a dozen or so prompts, so value depends on the content; people who often write character prompts can take a look, but those wanting a tool will find it too light. #Prompting#AI Agents#LLMs#Web Frontend

    Pill · One-time prompt fragments for AI companions: pick one manually, it applies only to the next reply and burns after use. Zero-dependency Web Component + 13 prescriptions.

    ceshihaox-dotcomJavaScript★22

  • C commander-mangen Automatically generates man pages including subcommands from commander.js programs, solving the real pain of maintaining CLI documentation and working out of the box. But the ecosystem already has commander's self-documentation and oclif, and with a narrow feature set this small tool could be quickly absorbed upstream rather than grow independently, limiting commercial room. #CLI#Documents#Build#commander-js

    Generate man pages from a commander.js program, including every subcommand's options.

    kofiadeyemiqTypeScript★33

  • C consulting-html-ppt An Agent skill that lays content out as McKinsey/BCG-style HTML+SVG slides with ECharts charts, output ready to open. Such "let AI make PPT" skills have clustered lately; its only difference is template style, and the pile of selling points in the description actually dilutes credibility. #Agent Skills#Presentations#echarts

    Consulting-firm-style HTML + SVG presentation page generation skill (Agent Skill) · McKinsey/BCG-style PPT · three-form determination + dual template library + ECharts

    sharptoolboxHTML★11

  • C magaza-mcp An MCP server connecting both App Store Connect and Google Play, solving the repetitive dual-platform data queries of mobile developers—a practical scenario. But serving only Turkish-speaking users severely limits the audience, and multiple dual-store MCP implementations already exist, making it hard to become an international standard dependency, with a low commercial ceiling. #MCP#iOS#Android#APIs

    Tek MCP sunucusu, iki mağaza: App Store Connect + Google Play. Tamamen Türkçe.

    tunaarikayaTypeScript★59

  • C UniversalModBridge A bridge layer meant to let old Minecraft mods run on new versions, a problem that has long existed in the modding community, and solutions like Sinytra Connector have not fully solved it. But the project calls itself WIP with no compatibility scope, so installing it now will likely mean hitting pitfalls. #Games#Game Dev#minecraft

    This is a WIP mod that enables the use of older mods in newer versions of minecraft

    gato001k1Java★9

  • C dotnet-memshell A technical sample demonstrating three .NET memory-shellcode injection points: HttpModule pipeline, WebSocket takeover, and Remoting URI registration. Valuable for red teamers and defenders building detections, but it is an attack-side tool that ordinary developers won't need and should treat with caution. #Exploits#App Security#Windows

    Three .NET memory shells at three distinct insertion positions - HttpModule pipeline, WebSocket connection takeover, Remoting URI registration. Zero impact on unmarked traffic, no new port, no web.con…

    lupingQAQC#★9

  • C BigNaiWa A pure front-end synthesis mini-game with self-written PBD physics, collision boxes generated from image outlines, zero dependencies and zero build, and honestly documented engineering details. A good Web physics learning sample, but the gameplay itself offers nothing new. #Games#Web Frontend#Animation

    Merge Big Milk Baby · pure front-end mini-game (self-built PBD physics + collision boxes based on image contours + TinyWebDB online leaderboard), zero dependencies, zero build

    YHSomeJavaScript★9

  • C humanbench A meme test webpage where you play an LLM, with the full creation process and retrospective. Worth a look once as a prompt novelty; more fun than useful, and it solves no real problem. #LLMs#Benchmarks#Web Frontend

    How many B are you? How many B are you? A meme quiz where you play the LLM, open-sourced with the full creative trail: every prompt, briefs, and a 3-part retro.

    alex-ybuildJavaScript★9

  • C duckapi Uses Playwright to drive a real Chrome and wraps Duck.ai as an OpenAI- and Anthropic-compatible interface, technically feasible. But it relies on automation to bypass server-side usage limits, can be blocked at any time, and cannot be a production dependency; there are already many similar free proxies, so it is not worth investing in. #LLMs#APIs#Automation

    OpenAI- and Anthropic-compatible API for Duck.ai, driven by a real Chrome browser via Playwright. Streaming chat, image generation, session reuse, and a web dashboard.

    hirotomasatoPython★14

  • C re-sseol-ch An Agent skill that rewrites papers into Korean community "음슴체" info posts, with a clear purpose and an honest tone. But the audience is basically Korean community writers, and it is essentially a prompt plus format template, so generality and reuse value are limited. #Agent Skills#Prompting#Content Writing#Learning

    An agent skill that explains papers as community info posts in the eumseum-che style

    subiniumPython★11

  • C OLY-Lang A compiled teaching language for Malagasy written in Rust, filling a tooling gap for a language with almost no native programming vocabulary; the motivation is clean and the scope restrained. But the audience is extremely narrow, almost no real project would use it today, and it is closer to an honest language design exercise. #Learning#Build#CLI

    OLY — A 100% Malagasy programming language designed to make programming easier to learn. Compiled, educational, and open source.

    tantelyorionRust★10

  • C qt-swift Qt bindings for Swift, filling a genuinely under-explored gap and appealing to Swift developers who want Qt on macOS. But it is only one line, "Qt for Swift," with no word on module coverage, build steps or how usable it is, so it looks like a just-started experiment. #SDKs#macOS#Build

    Qt for Swift

    torarnvSwift★8

  • C island Turns the MacBook notch into a small panel for music, files and meetings, emphasizing no account, no analytics and data staying local, which is well judged. But notch tools like boring.notch already exist as mature open source, and it never states its difference, so it is just another one. #macOS#Productivity#UI Design

    Island turns the notch on your MacBook into a place for music, files, meetings and more. Hover the notch to open it. No account, no analytics, nothing leaves your Mac.

    uzarixxSwift★8

  • C bluewake A native port of The Legend of Zelda: The Wind Waker to iOS/iPadOS, a substantial reverse-engineering and porting effort with pull in the retro-porting community. Nintendo's IP brings legal takedown risk, and a single-game port cannot be productized, so commercialization and becoming a standard dependency are essentially hopeless; it is purely a technical showcase. #Game Porting#iOS#Games#Reverse Engineering#zelda

    The Legend of Zelda: The Wind Waker, running natively on iPhone and iPad.

    chrissotraidisC★45

  • C kantor-agent A plugin that shows what Claude Code and its sub-agents are doing in a 3D office, described in Indonesian, with a clever visual. Agent visualization has seen several entrants this half year, and functionally it is more a toy; it has some observational value for debugging multi-agent flows, but not enough to be a tool. #AI Agents#3D#Plugins#claude-code

    Kantor Agent: kantor 3D tanpa konfigurasi yang menampilkan Claude Code & subagent-nya bekerja (plugin Claude Code)

    humaedihumeJavaScript★29

  • C bilibili-ambilight A browser extension that adds ambient lighting to the Bilibili player, openly adapted from a YouTube version. For people who often watch anime on Bilibili it is a low-cost improvement, but originality is low and alternatives already exist; it is a nice little extra, not worth ten minutes specifically. #Browser Extensions#Video#UI Design#bilibili

    Bilibili Ambient Light: a browser extension that adds ambient light around Bilibili videos (unofficial, adapted from Ambient light for YouTube)

    iceorange-devJavaScript★16

  • C composimplex Combinatorial LLM decoding on the probability simplex, with paper code. It's concrete work for decoding-strategy researchers, but still research code, lacking docs and baseline comparisons, so the bar for direct engineering use is high. #LLMs#ML Research#Benchmarks

    Composable LLM decoding on the probability simplex: library and paper code.

    KickItLikeShikaPython★12

  • C Dungeons2_linux_fix A Linux run fix for Dungeon 2, and 36 stars in half a day shows people really were stuck. But the repo doesn't say which Proton/Wine version the patch targets or what it changes, so readers can only gamble; similar fixes usually live in ProtonDB comments. #Linux#Games#Performance

    Kubas556C★36

  • C PyGroup3r A Python rewrite of the Windows privilege-escalation enumeration tool Group3r, lowering the barrier to AD security testing and useful for penetration testers. A well-known security account gave it initial stars, but it is a port of an existing tool with no innovation, commercialization is limited because red-team tooling is hard to license, and it will not become a new standard dependency. #App Security#Exploits#Windows

    Group3r but in Python

    S3cur3Th1sSh1tPython★38

  • C VLCoT A research implementation on stage-verifiable latent reasoning, involving observational verification and dependency-aware local repair. The method is academic, with no usage instructions or reproduction scripts, so ordinary readers can't judge its quality. #ML Research#LLMs

    VLCoT research implementation: stage-verifiable latent reasoning, observation-based verification, and dependency-aware local repair.

    apromisedlandPython★32

  • C video-illustrator An agent skill that turns narration and assets into short films, pitched on using your own voice and footage. It heavily overlaps with several video skills in the same batch and lacks sample films or failure-boundary documentation. #Agent Skills#Video#AI

    Your voice, your assets, any look you choose — an Agent Skill that turns your narration and real assets into a short film.

    axtonliuJavaScript★12

  • C ai-is-fast-your-tests-are-slow A headline-style opinion piece plus tutorial, along with an agent skill that makes tests faster, by a well-known Korean AI educator. The content sits between article, research and tool, with a vague positioning; those just looking for something runnable will not get it from this repository. #Testing#Performance#Agent Skills#Learning

    AI is fast. Your tests are slow. Research, a tutorial, and an agent skill for making test suites fast — in the AI coding era the bottleneck is testing.

    hunkimPython★9

  • C glm53-tensorfold-spark Runs GLM-5.3-Flash (abliterated EXL3) on two DGX Sparks with the TensorFold engine, claiming 1.8x decoding speedup, a custom inference-optimization experiment. The author marks it WIP, and it depends on a niche hardware combination and an unofficial quantized model, making it hard to reuse with no standard dependency or business path. #LLMs#GPU#Performance

    GLM-5.3-Flash (abliterated EXL3) on 2x NVIDIA DGX Spark with the TensorFold engine: 1.8x faster decode than vLLM, 4x256k concurrent threads, byte-exact speculative decoding. Work in progress.

    jayleatonPython★38

  • C SiCo Wraps Codex CLI as an integrated circuit design assistant, a more specific direction than yet another general coding assistant. But the description stops at "platform intelligent assistant," never showing which chip-engineering step it saves or offering a usage example; among many Codex wrappers, this one just swaps in an industry shell. #AI Agents#CLI#Hardware#LLMs

    SiCo (矽科) — a custom integrated circuit design platform assistant built on Codex CLI

    opencad-abuPython★11

  • C nd-video-studio Lays out explainer videos in HTML, then adds AI narration, music and mixing to output MP4, packaged as a Claude Code / Codex skill. The pipeline idea is clear, but it is deeply tied to NeuralDeep's own models, the description shows the brand more than usage, and it is basically unusable without that service. #Video#AI#Agent Skills#Speech

    Explainer videos from HTML (HyperFrames → MP4) with AI narrator, music and mixing — Claude Code / Codex skills on NeuralDeep models

    vakovalskiiPython★11

  • C TabAlign The official implementation of a table-reasoning paper, claimed to be published at NeurIPS 2026, currently with only plans and evaluation code and no reproducible steps. It has reference value for researchers in the same area, but is useless to ordinary developers. #ML Research#LLMs#Benchmarks

    Official code for "From Table to Cell: Attention for Better Reasoning with TabAlign" (NeurIPS 2026)

    ThomasK1018Python★9

  • C starfox-enhanced A portable build based on the Star Fox decompiled source, labeled as AI-assisted. It is valuable to retro porting enthusiasts, but such projects usually get stuck on assets and legal gray areas, and the repository does not say whether it bundles resources or needs an original ROM, so practical usability is doubtful. #Games#Game Porting#3D#star-fox

    Star Fox Enhanced source and portable platform builds (Assisted by Codex AI)

    kandowontu2C++★12

  • C intune-docs-mirror Mirrors Microsoft Intune and Autopilot documentation into a repo where each commit maps to one change, making edits easy to track. Somewhat useful for people managing many Windows devices, but it is essentially a content mirror rather than a tool, so it scores low under the rules. #Documents#DevOps#Automation#Windows

    Intune and Autopilot docs mirrored from Microsoft Learn, one commit per change

    merillLanguage unknown★12

  • C po0-clash A proxy client built on FlClash that adds po0 firewall auto-whitelisting, saving domestic users who need split-routing rules from manual upkeep. But it is a re-wrap rather than a new implementation, differing by one whitelist logic, and nobody explains the long-term cost of tracking upstream updates. #Networking#Privacy#Android

    po0-clash: FlClash-based proxy client with built-in po0 firewall auto-whitelist (po0fw)

    yuuuki-creationDart★14

  • C EternalVR An unofficial VR mod for DOOM Eternal with OpenXR, 6DoF, and motion controls, addressing a real demand among VR players. But another nearly identically named Doom Eternal VR project appeared the same day, making completeness and provenance hard to verify, and setup is demanding. #Games#3D#Game Dev

    EternalVR: a free VR mod for DOOM Eternal (OpenXR, stereo, 6DoF, motion controls). Unofficial fan project.

    ArcadaLabs-JasonC++★12

  • C BusterBustsLoose A native Windows port of the SNES Tiny Toon adventure, with widescreen, HUD rearrangement and level select, and it honestly requires your own US ROM. The port has technical merit, but the audience is nostalgic players, and it's unclear whether it's a decompilation or an emulator wrapper. #Games#Game Porting#Windows

    Windows PC port of Tiny Toon Adventures: Buster Busts Loose! (SNES) with true widescreen, HUD re-anchoring, display options, cheats, level select and music player. Requires your own USA ROM.

    DarhkfoxC★14

  • C motion-showreel A 15-second motion piece drawn in real time with Jetpack Compose and AGSL shaders, somewhat interesting as a showcase of Android graphics capability. But it is an artwork, not a tool, so others cannot take it away; those wanting to learn Compose shaders can read the source, while other readers gain little. #Animation#Android#UI Design#jetpack-compose

    A 15-second motion-design showreel drawn live in Jetpack Compose + AGSL shaders. Native Android, no keyframes. Built with Claude Opus 5.5.

    c5incoKotlin★16

  • C TIRx-kernels A GPU kernel collection accompanying TIRx, essentially an accessory to the previous repo. On its own it is just kernel samples with no word on which hardware it targets or how to benchmark it, so it is of limited value without the harness. #GPU#ML Research#Performance

    Curated high-performance TIRx GPU kernels

    mlc-aiPython★8

  • C youtube-transcript-api Wraps YouTube subtitle fetching into a REST API that needs no Google API key, with search, channel and playlist endpoints and an OpenAPI description, so integration is cheap. But mature libraries already exist, and it says nothing about stability or compliance boundaries, making it a repackaged duplicate implementation. #APIs#Downloaders#Automation

    YouTube Transcript API: get YouTube video transcripts via REST API, no Google API key. Plus YouTube search, channel and playlist endpoints, OpenAPI spec and examples.

    tubeagentkitLanguage unknown★14

  • C clearwater-vrc A port of clearwater's shallow coastal scene into VRChat, providing wadeable shallows and coastal assets. Ready-made material for VRChat world creators, but it's just an asset port with nothing new technically or tool-wise. #3D#Game Dev#Game Porting

    Clearwater VRC: clear, walkable shallows and coasts for VRChat worlds (a port of clearwater by Aureliengmz)

    bmbb93C#★11

  • C math-drill An unofficial English localization of the Japanese elementary-school arithmetic drill Dopa Drill, usable right in the browser, practical for parents and teachers. Low originality is stated openly; the question is whether the question bank's source and licensing hold up, and there are already many similar free practice sites. #Learning#Web Frontend#Productivity

    Unofficial English localization of Dopa Drill (ドパドリル): a browser math drill for grades 1–6

    elatdJavaScript★9

  • C webgpt-api Wraps ChatGPT web sessions into a locally listening, OpenAI-compatible API, tempting for anyone wanting to save on API fees. It is essentially web reverse engineering to bypass official billing, will break whenever the other side changes, and suits no production scenario; many clones have existed for years. #LLMs#APIs#Reverse Engineering

    Turns the ChatGPT web Chat mode into an OpenAI-compatible API that listens only on localhost.

    ulfmawJavaScript★14

  • C genpark-alpaca-sharegpt-format-converter-skill Converts between Alpaca and ShareGPT dataset formats with schema validation, occasionally useful for people preparing training data. The functionality is narrow and ready-made scripts abound; the same author's batch of genpark-* repos released within days look like placeholders, making maintenance intent hard to gauge. #LLMs#Agent Skills#Data Analysis

    Bidirectional dataset format converter and schema validator between Alpaca and ShareGPT formats

    Alpha-ParkPython★7

  • C hearthlight.github.io A pixel-art adventure demo in the browser, with a ten-chapter story, eight characters, and phones usable as controllers for 1–8 player co-op. The phone-as-controller party mode is a clever touch, but overall it's a personal showcase with no reusable library or docs. #Games#Web Frontend#UI Design

    A cozy pixel-art adventure in the browser (a demo): a story in ten chapters, eight heroes, Party Mode for 1–8 players with phones as controllers. Everything procedural.

    HearthlightJavaScript★12

  • C video-processing-pipeline A Spring Boot async transcription service: S3 presigned uploads, SQS events, an idempotent worker with leases and retries, tested with LocalStack and Testcontainers. The architecture is clearly explained, but it's more a practice reference implementation with no real-load validation. #Speech#Video#DevOps#Testing

    Spring Boot service that transcribes uploads asynchronously: S3 presigned upload → S3 event → SQS → idempotent worker with DynamoDB leases, retries and a DLQ. Tested against LocalStack with Testcontai…

    LindseyZ1205Java★14

  • C video-skills A set of video-production skills for Claude Code and Codex, essentially a collection of prompts and workflows. No code or verifiable output, quality depends entirely on the author's taste, and there are already many similar skill repos. #Agent Skills#Video#Prompting#LLMs

    Reusable video production skills for Claude Code and Codex

    adamperlisLanguage unknown★14

  • C fox-floe A minimalist rowing mini-game built with Three.js, playable offline, with art made in Blender and a restrained scope. A nice weekend project, but the gameplay is thin and no different from countless browser mini-games. #Games#Web Frontend#3D

    A tiny endless river game: a fox rows down icy rapids. Three.js, Blender-made assets, fully offline.

    atmirrrJavaScript★10

  • C SpatialSpeak-VLM Looks like the companion repo for a spatial reasoning paper, whose title mentions QA-native 3D reconstruction using local and global context. The topic tracks research trends, but the repo contains only a title, with no data, evaluation or reproduction instructions, so there is no way to judge whether the conclusions hold. #ML Research#Multimodal#LLMs

    SpatialSpeak: QA-Native Reconstruction with Local and Global Context for Spatial Chain-of-Thought Reasoning

    yangcaoaiLanguage unknown★19

  • C CVE-2026-94545 The title only says "Next.js RCE", with no affected versions, reproduction steps or fix guidance. If true it would be a high-profile vulnerability, but in this form it can't be verified, so security teams shouldn't act on it; wait for a version with PoC details and patch information. #Exploits#App Security#Network Security

    Next.js RCE

    EQSTLabPython★13

  • C WorldPlay2 Claims to extend real-time interactive world models in control and temporal range, but the repo has no language tag, usage, paper, or demo link, and its stars seem inexplicable. World models are hot, but this repo currently looks like a placeholder. #ML Research#Video#Multimodal#AI

    WorldPlay2: Extending Real-Time Interactive World Models in Control and Horizon

    WorldPlay2Language unknown★32

  • C one-prompt-video-skills Claims to generate a complete talking-head video from one sentence, chaining scriptwriting, digital humans, B-roll and compositing into a series of Claude Code Skills. The pipeline sounds complete, but every step relies on external services, and the repository is more process orchestration plus prompt collection; final video quality cannot be judged from the description. #Agent Skills#Video#LLMs#Prompting

    Generate a complete talking-head video in one sentence: Claude Code Skills for scriptwriting, digital-human presenter, B-roll, HyperFrames compositing, and frame-extraction checks

    huangbai-AIJavaScript★25

  • C duomind Wants to pair a local small model with a model called Jev for fast-and-slow dual-system inference, claiming it is faster and more accurate and compatible with the OpenAI interface. The core components have no traceable source and the metrics cannot be verified, reading more like concept packaging than a reproducible solution. #LLMs#Performance#AI Agents

    When Jev meets LLM -- Pair a small local LLM (System 2) with Jev model (System 1) to make small model more faster and accurate — OpenAI-compatible, private, and almost free to run in Kilo code and Cli…

    j1s4nnPython★16

  • C data-visualizer A small Portuguese-language tool that walks through Python data structure changes step by step in the terminal, showing beginners how linked lists and stacks move. The idea is common and similar visual teaching aids already exist; usable as a teaching aid, but narrow in audience and lacking differentiation. #Learning#Data Analysis#Terminal

    Esquema de visualização passo a passo de estruturas de dados em Python no terminal.

    malu-zinhaHTML★12

  • C interfaces-that-think A set of UI interaction prototypes, one experiment per folder, exploring new ways of operating. Fine to browse as an inspiration library, but with no reusable components and no conclusions it is a personal experiment ground rather than a tool, and readers leave with little. #UI Design#Web Frontend#Animation

    UI/UX experiments for new interactions, with one folder per prototype.

    onepixelawayJavaScript★9

  • C genpark-locality-sensitive-hashing-lsh-cosine-index-skill Random hyperplane LSH for cosine similarity candidate recall is a mature classic retrieval algorithm, and packaging it as a callable skill is straightforward. The problem is it's just a thin algorithm wrapper with no measured data scale or recall rates; those needing approximate nearest neighbors will mostly use faiss or hnswlib directly. #LLMs#RAG#Agent Skills

    Random hyperplane Locality-Sensitive Hashing (LSH) for sub-linear cosine similarity candidate generation

    Alpha-ParkPython★7

  • C KHARVOX-ARGENT Another DOOM Eternal VR conversion mod, colliding with the EternalVR that appeared the same day, with only a one-sentence description. VR players might benefit, but there's no telling which is more complete or installable; the duplication itself shows this ecosystem is still very early. #Games#3D#Game Dev

    A Doom Eternal VR conversion mod

    CactusVRStudiosC++★17

  • C skills A set of skills that help agents extract a design system from an existing app and migrate each screen over. The direction is practical, but for now it's just a skill collection with no sample project or before/after comparison. #Agent Skills#UI Design#Web Frontend

    Agent skills for building a design system from your app and moving every screen onto it.

    arla6kaJavaScript★8

  • C omarchy-starship-to-orbit-theme A Starship launch theme for Omarchy, with white text, liquid glass and Raptor exhaust, a very clear aesthetic direction. It is just a color scheme plus wallpaper pack; those who like it install it in a minute, and those who do not are unaffected; it addresses no new problem and has no engineering value. #Linux#UI Design#hyprland#omarchy

    An Omarchy theme celebrating Starship's flight to orbit: white text on lightly frosted liquid glass over the launch, with the Raptor plume running through it all.

    ejuroLua★11

  • C DN-MOPD An improved multi-teacher online distillation method with domain normalization, plus code for the same-named paper. It's very niche research work with unknown code usability, and basically unusable except by researchers in the same direction. #LLMs#ML Research#Benchmarks

    Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation (DN-MOPD)

    LiXin97Python★10

  • C AirSIM Claims to do cross-device calls and messages, covering Samsung Android, iPhone, Apple Watch, Linux and Workers, a scope so broad it is suspicious, and it also clashes with Microsoft's AirSim name. A non-commercial license with such broad promises makes it even more necessary to see something runnable first; right now there is only a description. #Social#Self-Hosted#iOS#Android

    Source-available, noncommercial cross-device calling and messaging for Samsung Android, iPhone, Apple Watch, AVF Linux and Cloudflare Workers.

    kai-wu-cortexSwift★17

  • C FreeOSINTUI Marketed as a free alternative to osint-ui, with a description that's just one line of links. No feature comparison and no screenshots; the value of an open-source alternative depends on how much of the original it covers, which is currently invisible. #Open-Source Alternatives#Network Security#Privacy#osint-ui

    Free alternative to https://osint-ui.com

    KerloooJavaScript★10

  • C motion-graphics-skills A motion graphics skill pack for Claude Code and Codex, claiming it can produce brand films and kinetic typography. The description is mostly marketing language, with no sample output or failure cases, so it is unclear how it beats writing prompts directly. Such skill collections have been appearing in clusters lately; treat with caution. #Agent Skills#Animation#Video

    Motion graphics for AI agents. Creative direction + production skills for Claude Code & Codex, powered by HyperFrames. Build brand films, kinetic type, 2D/3D and programmatic video. Every frame is cod…

    imMamdouhaboammarJavaScript★9

  • C shanghaitech-elrc-study A local video player serving only one course, ELRC at ShanghaiTech, and students still have to find a way to download the course media themselves. The scope is clear, but it's extremely narrow and carries a gray-area whiff of bypassing the course platform, with no value to most developers. #Video#Learning#Downloaders

    Local video-first ShanghaiTech ELRC study player; students download their own course media

    OwenxuSHPython★10

  • C SolveEdit Research code called SolveEdit that only mentions plan and evaluation parts, without saying what problem it solves or how to run it. It has some stars but near-zero information, so its value can't be assessed. #ML Research#AI

    SolveEdit-Plan and evaluation code for SolveEdit

    WenjieShuPython★15

  • C bgfx-dlss5-nr Claims to port DLSS5 neural rendering to BGFX, but DLSS is closed-source and there's no public DLSS5, and the repo has no code description or build steps. Either a placeholder or clickbait—wait for substance. #Game Dev#GPU#Performance

    DLSS5 Neural Rendering ported to BGFX

    andrewmd5Language unknown★8

  • C genpark-messaging-agent-proactive-nudge-scheduler-skill Proactive reminder scheduling for messaging personal agents, with urgency decay and do-not-disturb filtering, is a concrete idea. But it doesn't say which messaging platform it integrates with or how it's triggered, so no runnable form is apparent; people wanting this would likely write a few dozen lines of scheduling logic themselves. #AI Agents#Agent Skills#Productivity

    Proactive contextual nudge scheduler for messaging-native personal agents with urgency decay and quiet-hours filtering

    Alpha-ParkPython★8

  • C REPI A representation injection method for diffusion Transformers, with only a paper-style title and one line saying "Code coming soon". Researchers in the same area can wait for the code; there's nothing runnable now. #ML Research#Image Generation#AI

    Scaffold Then Internalize: Representation Injection for Diffusion Transformers. Code coming soon.

    JeneveuxpasCSS★9

  • C piggery Aims to organize Pi, Claude Code and Codex coding agents into a team with divided labor. Multi-agent orchestration is already crowded, and its description is almost all pigpen metaphors and emoji, never saying how tasks are split or conflicts merged, so after reading you still don't know why it beats a single agent. #AI Agents#LLMs#CLI#Automation

    Piggery 🐖 - Lợn cày tasks: Your coding agents are pigs, piggery is the farm. Pi, Claude Code and Codex agents plowing tasks as one team.

    sting8kGo★10

  • C awesome-Fonts-skill A list of 94 Chinese fonts classified by glyph shape, plus a font-picking skill. Handy for people stuck choosing fonts when typesetting, but it is still a resource collection, manually curated, with no verifiable selection logic, not much different from font navigation sites online. #Agent Skills#UI Design#Documents

    The Beauty of Glyphs — character charm and spirit within a square inch. 94 Chinese fonts and variants, classified by glyph shape, with specimens and sources. Browse online, or use as a Skill to choose…

    tanglele110-hashHTML★8

  • C fomo-cc An Agent skill that mocks AI hype and hallucination then offers a "way out," an amusing stance that mirrors hype-chasers back at themselves. But it is essentially a prompt roleplay, read once and done, with little reuse beyond emotional value. #Agent Skills#AI Agents#Prompting

    An Agent Skill that roasts AI-hype delusions - then shows the way out.

    zack-the-workerLanguage unknown★17

DSkip99
  • D Qwen3.8-Flash-Next-Single-DGX-Spark-TensorFold A Shell deployment script for running Qwen3.8 Flash on a single DGX Spark, valuable for giving a reproducible edge/single-machine inference configuration. It is useful reference for teams holding that hardware, but it is essentially configuration glue with no algorithmic or architectural innovation, so it is unlikely to become a standard dependency and has little commercialization room. #LLMs#GPU#DevOps

    Qwen3.8 Flash Next on one DGX Spark (TensorFold)

    MiaAI-LabShell★83

  • D deepseek-whale-girl A fan-art resource repository around DeepSeek's whale-girl mascot, with reference images and prompts, riding IP hype for short-term stars. It is fan-content aggregation rather than a technical asset with standard-dependency value, and commercialization is doubly constrained by IP copyright and the character's hype cycle, making retention hard once interest fades. #Image Generation#Prompting#Content Writing

    A community-maintained fan resource library for the whale girl, organizing character reference images, generation prompts, community works, and related projects to provide creators with convenient ref…

    TreapGoGoCSS★137

  • D museai An OpenAI-compatible gateway for the muse.ai personal agent wired into 9Router multi-model routing, essentially API adaptation glue. It is immediately usable for muse.ai users, but the gateway/routing space already has strong players like LiteLLM and OpenRouter, and without core technical barriers it is unlikely to become a standard dependency. #APIs#SDKs#AI

    High-performance OpenAI-compatible API gateway and bridge for muse.ai personal AI agents, ready for 9Router multi-model integration.

    d4ncbozPython★39

  • D electromagnetism An interactive electromagnetic-field exploration web page generated by Claude, suited to building physics-teaching intuition. The educational value is real but it is a one-off content product without a deep data or simulation core, hard to accumulate into a standard teaching library, and mature resources like PhET already occupy the space, leaving insufficient differentiation. #Learning#Web Frontend#Science

    An interactive Electromagnetic field exploration built using Claude Opus 5.5

    thebuggeddevJavaScript★42

  • D Twitch-AIO An all-in-one tool for bulk-registering Twitch accounts and managing follows, chat and tokens, squarely in the social-media automation gray market. There is some real user demand driving stars, but it violates platform terms, accounts are easily banned, and it cannot be commercialized or enter legitimate dependency chains, with legal and compliance risk as the hard flaw. #Social#Automation#Scams

    All-in-one Twitch automation toolkit - account creation, follows, chat, and token handling in one fast, reliable Python project.

    RishabhEvansPython★83

  • D nesterm Runs NES games as ASCII art in the terminal, a retro geek toy whose 47 stars come from retro-community sharing. Technically it combines an emulator with character rendering, solves no novel problem, has no commercial or standard-dependency path, and its popularity will fade with the novelty. #Terminal#CLI#Games#nes

    Play NES games as ASCII art in your terminal

    kathocJavaScript★47

  • D abstract-algebra-promo Canvas/WebGL animations of abstract algebra concepts with procedural soundtracks, a math-visualization art demo. It has some teaching and viewing value but no interaction depth or course system, so it is hard to turn into reusable teaching components; the 68 stars mostly reflect visual novelty and there is almost no monetization path. #Animation#Audio#Learning

    Abstract Algebra: The Beauty of Structure | Canvas/WebGL animation and procedural soundtrack

    AnctyEnly453HTML★68

  • D antigravity-fixer A credential-cleanup fix script for the Google Antigravity editor's 'not eligible' error, a typical emergency tool for a hot product that gains stars from short-term search traffic. It will inevitably break once the official fix lands, has no accumulating value, and is not a dependable component, making it a one-off hype product. #Debugging#CLI

    Fix Antigravity 'not eligible' error - diagnose, clean credentials, auto-fix

    fulldiagnosePython★108

  • D FlipperAmiiboCollection NFC dumps of over 900 Amiibo, neatly organized and deduplicated, ready to use for Flipper Zero owners. But this is Nintendo's commercial data, so distribution carries copyright risk; it's a resource repo, not a tool. #Piracy#Games#Hardware

    900+ ready-to-use Amiibo NFC files for Flipper Zero, covering nearly the entire Nintendo Switch Amiibo catalog — cleaned, deduplicated, organized and NTAG215-fixed.

    Gutierre0x80Language unknown★24

  • D ricoh-gr4-firmware-analysis-and-feature-expansion From the repo name, firmware analysis and feature extension for the Ricoh GR4 camera, which is interesting to the niche camera-hacking crowd. But there is no explanatory text at all: no word on what was analyzed, whether it is reusable, or where the risks lie. Firmware reverse engineering should spell out its boundaries, and this fails that test. #Reverse Engineering#Hardware#Embedded

    radium-wangPython★18

  • D dsh-550c-boot A 550C boot animation plugin for DeepSeek Harness that hides the host loading card and unifies terminal colors. A small toy for the few who care about boot ceremony, but its use is extremely narrow, and the description stacks jargon so densely it takes two reads to see which layer it changes. #Plugins#Animation#Terminal#UI Design

    550C boot animation for DeepSeek Harness — the first frame is injected by the host's half, so DSH's Loading card never shows its face; native desktop title bar buttons are repurposed into terminal col…

    yannicksong0106JavaScript★20

  • D cute-chat-stickers A sticker and kaomoji drawer for two-person chats, with emoji backgrounds and bubble reactions, is a small frontend toy. It targets web chat scenarios wanting playful interaction, but the description is full of decorative symbols with no install instructions, and it's unclear how it beats existing emoji libraries. #Web Frontend#UI Design#Social

    ♡ press anywhere, stick anything · emoji background, bubble reactions, sticker shelf & kaomoji drawer for chats between two

    Anko3oJavaScript★17

  • D background-agents The description is just "background agents", so it's unclear whether it's a framework, template or demo. The author's fame may bring initial stars, but there's no evaluable information for readers. #AI Agents#Automation#LLMs

    background agents

    HendrixerTypeScript★18

  • D genpark-adaptive-retrieval-router-skill One of a batch of skill repos released by the same org within two days, wrapping RAG routing as a classifier. No benchmark, no install instructions, and the description looks template-generated; okay alone, but hard to take seriously in bulk. #RAG#Agent Skills#LLMs

    Adaptive RAG retrieval routing classifier directing queries across direct LLM, single-hop, and multi-hop RAG

    alphaparkincPython★7

  • D genpark-agent-human-in-the-loop-approval-gate-skill One of the same batch, providing an agent approval gate and risk grading. The concept is on the right track, but there are no policy examples and no explanation of how to integrate with any agent framework—it reads like jargon stacking. #AI Agents#Agent Skills

    Human-in-the-Loop policy gatekeeper enforcing risk tiers, reversible checkpoints & token approval bounds for autonomous agents.

    alphaparkincPython★7

  • D meeting-preparation-agent A meeting-preparation agent whose description is all vague talk about solving real problems and improving accessibility, with no technical detail or differentiation, a typical generic LlamaHackathon-style submission. Meeting-notes and preparation tools are plentiful, it shows no real user signal, and both commercial and dependency value are very low. #AI Agents#Productivity

    Our project is a technology-based solution designed to address a real-world problem in a simple and efficient way. It focuses on understanding user needs, improving accessibility, and providing a prac…

    paritalamounika07-designPython★44

  • D MojoRecomp Claimed to be a native PC recompilation project for a game from the "Crash Titans era", but the description is so vague you can't tell which game or how far along it is. Game recompilation has an extremely high bar, and this repo currently has no verifiable output. #Game Porting#Reverse Engineering#Games

    A native PC recompilation project for the Crash Titans era.

    OAleexC++★8

  • D genpark-adsr-envelope-wavetable-synth-skill One of the same batch, a Python implementation of ADSR envelopes and wavetable synthesis. Textbook-level code, but with no audio samples or integration method, so it's impossible to judge whether it's usable. #Audio#Agent Skills

    ADSR envelope state machine and band-limited wavetable synthesizer for tone and waveform synthesis.

    alphaparkincPython★7

  • D genpark-quadratic-programming-active-set-skill One of the same batch, solving constrained quadratic programming with active-set/KKT. The algorithm itself is easy to verify, but there are no tests and no comparison with scipy or cvxopt, making usability hard to judge. #Science#Agent Skills

    Active-set Karush-Kuhn-Tucker (KKT) quadratic programming (QP) solver for quadratic costs with equality/inequality constraints

    alphaparkincPython★7

  • D awesome-tauri-apps Yet another awesome list; 17 stars half a day after creation proves nothing. Lists do not solve specific problems, and value depends entirely on whether the maintainer keeps curating long term, while most such repos stop updating within weeks. Those looking for Tauri apps are better off checking the official showcase. #Web Frontend#Productivity#tauri

    Curated list of desktop and mobile apps built with Tauri.

    chrysler-stoneLanguage unknown★17

  • D tarski Claims to compress three forward passes into one on diffusion language models while improving results, with model names of unknown origin mixed into the terminology. It lacks experimental setup and reproducible scripts, and the metrics cannot be checked, so it reads more like packaged research notes and is not credible for now. #ML Research#LLMs#Performance

    Reading diffusion language models for decisions: same OpenJev model, one pass instead of 3.65, +5 points, calibrated, no training. Plus cheap task branches on a frozen encoder.

    looskisPython★11

  • D seis-AI Seismic waveform P/S phase detection with uncertainty estimation, built on SeisBench and deep sequence models, is a more concrete topic than the author's other repos. But it's all vision-style description with no dataset, metrics, or run instructions; people doing this work would mostly just use SeisBench's official pipeline. #ML Research#Science#Data Analysis

    SEIS AI is an AI-powered earthquake detection and seismic analysis platform that processes seismic waveforms to detect P/S phases, estimate event uncertainty, and improve earthquake detection under re…

    Anirudh10-pngLanguage unknown★21

  • D shanghaitech-kimi-gateway Builds a local gateway for Kimi Code and auto-configures an IDS passkey, which reads like circumventing official login or quota limits. The description is vague about applicable scenarios and compliance boundaries, so no one but the author can safely reuse it. #LLMs#APIs#Secrets

    Local Kimi-K3 gateway and automatic IDS passkey setup for Kimi Code

    OwenxuSHPython★9

  • D genpark-black-scholes-merton-greeks-engine-skill One of the same batch, a Python implementation of Black-Scholes pricing and Greeks. Standard formulas and little code, already built into any quant library, and it lacks examples and numerical error discussion. #Fintech#Agent Skills

    Black-Scholes-Merton European option analytical pricing and first/second-order Greeks (Delta, Gamma, Vega, Theta, Rho) engine.

    alphaparkincPython★7

  • D genpark-half-edge-mesh-dcel-topology-skill One of the same batch, implementing half-edge/DCEL mesh topology and Euler characteristic. A common structure in geometry processing, but it isn't connected to any modeling or rendering pipeline, so a standalone library has limited meaning. #3D#Agent Skills

    Doubly Connected Edge List (DCEL) Half-Edge mesh topology with vertex incident cycles and Euler characteristic

    alphaparkincPython★7

  • D genpark-quantum-qubit-state-vector-simulator-skill One of the same batch, doing multi-qubit state-vector simulation. Fine as a teaching demo, but libraries like Qiskit already cover this, and it doesn't say how many qubits it can scale to. #Science#Agent Skills

    Multi-qubit complex state vector simulator calculating superposition amplitudes, state fidelity, and probability distributions.

    alphaparkincPython★7

  • D genpark-quaternion-rotation-slerp-skill One of the same batch, unit quaternion multiplication and SLERP interpolation. A basic implementation nearly every graphics library already includes, so packaging it as a skill adds no value. #3D#Agent Skills

    Unit quaternion 3D orientation engine with Hamiltonian multiplication, vector rotation, and SLERP interpolation.

    alphaparkincPython★7

  • D ZcodePro Claims to be the "official plugin marketplace" for some open-source build, but never explains what ZCode is or where the plugins come from. Branding it official without any provenance makes its credibility doubtful, so it is not recommended. #Plugins#Editors#LLMs

    Official ZCode plugin marketplace for the open-source build

    luxi233Python★10

  • D AutoPilotPM Claims to fully automate trading across Polymarket, Kalshi, Binance and on-chain DEXs with automatic risk control. The coverage is suspiciously broad, with no strategy explanation, backtest or risk disclosure; such "automatic money-making" repos offer developers almost no verifiable value, and the description alone is enough. #Crypto#Fintech#AI Agents#LLMs

    Open-source autonomous trading system powered by AI, built to monitor and trade across 1,000+ markets including Polymarket, Kalshi, Binance, Hyperliquid, Solana DEXs and multiple EVM networks. Finds o…

    recogardtechTypeScript★15

  • D model-guard A model monitoring platform using KS/PSI for drift detection plus a RAG layer for diagnostic suggestions. The features themselves are reasonable, but the description stops at a capability list with no integration method or threshold settings; mature options like Evidently already cover this space. #ML Research#Observability#Data Analysis

    Model Guardian is an AI-powered ML model monitoring platform that detects data drift, model performance degradation, and data-quality issues. It uses KS/PSI-based drift detection, performance metrics,…

    Anirudh10-pngLanguage unknown★21

  • D genpark-agent-self-consistency-consensus-scorer-skill One of the same batch, packaging multi-path chain-of-thought voting as a skill. Self-consistency scoring is already mature in papers and existing libraries, and here there's neither comparison nor usage instructions. #LLMs#Agent Skills

    Multi-path chain-of-thought self-consistency consensus scorer and majority voter aggregating divergent LLM reasoning trajectories into high-confidence deterministic outputs.

    alphaparkincPython★7

  • D genpark-percolator-distributed-transaction-protocol-skill One of the same batch, claiming to implement Percolator snapshot-isolation distributed transactions. This is a classic database protocol, and a single skill file can't support a usable implementation; dependencies and tests are also unaddressed. #Databases#Agent Skills

    Percolator snapshot-isolated distributed transaction protocol with primary-lock two-phase commit

    alphaparkincPython★7

  • D genpark-quantum-teleportation-entanglement-circuit-skill One of the same batch, generating Bell states for quantum teleportation demos. Textbook experiment content, missing circuit diagrams and not connected to a real simulator—little value as a skill pack. #Science#Agent Skills

    EPR Bell state generator creating maximally entangled qubit pairs (|Phi+>) for quantum communications and teleportation protocols.

    alphaparkincPython★7

  • D real-time-recommendation-analytics A recommendation analytics platform built on PySpark plus Kafka, a two-tower model and Qdrant, covering a full tech stack. It's more like a data-engineering portfolio piece: no data description, no deployment steps, hard to reproduce and not really usable. #Data Analysis#Databases#ML Research

    End-to-end retail analytics and recommendation platform using PySpark, Kafka, PostgreSQL, Power BI, PyTorch Two-Tower, Qdrant, FastAPI, Docker and Azure.

    Jiarui-0410Python★17

  • D PlayerDoll Player model rendering written in GLSL, claiming real-time skin updates. With no screenshots and no integration notes, it's unclear which engine or platform it runs on, leaving readers to guess. #Games#3D#GPU

    A shader-based player doll with real-time skin updates.

    Null-KGLSL★11

  • D mojito Claims to be a local, evolving personal AI benchmarked against cloud assistants, but it's all vision, with no installation method, architecture description or runnable demo. Personal AI narratives are already everywhere, and there's no evidence it does more than existing local assistants. #AI#LLMs#Privacy

    Mojito — the local, evolving alternative to cloud personal AI like Meta Muse.

    TimeLoverccTypeScript★10

  • D JARVIS-VORNEX A Windows voice desktop assistant that claims to operate safely in real apps, yet offers no demo, permission model, or install instructions—only a signature. A vague vision-type repo that can't be verified; skip it. #AI#Speech#AI Agents#Windows

    Voice-first desktop assistant for Windows — talk to your PC and it works inside real apps, safely. By VORNEX.

    VORNEX-01Python★9

  • D DiPlay-CN Only says "DiPlay Chinese localization," without explaining what the original project is, which parts are localized, or how it coexists with the original. It may save effort for users wanting a Chinese interface, but for developers the information is near zero, and localization distribution often raises licensing issues. #Android#Games#Plugins

    DiPlay Chinese localization

    serein-moriiKotlin★17

  • D ESP32-S3-Smart-Multi-Input-Control-System A demo project where an ESP32-S3 reads dual joysticks, a potentiometer, and a 16×2 LCD, suitable for embedded beginners copying the wiring. It's essentially coursework-level example code with no protocol description or firmware structure, making it hard to use as a starting point for a real control system. #Embedded#Hardware#IoT

    ESP32-S3 Smart Multi-Input Control System is an embedded control project featuring two analog joystick modules, a potentiometer, and a 16×2 I2C LCD. The ESP32-S3 reads real-time analog and digital inp…

    Ani579C++★10

  • D Inbox-pilot Yet another AI email assistant claiming to classify, summarize, and generate replies, with a full tech stack listed but no verifiable details or deployment instructions. The same author released several 'AI platform' repos in one day, looking more like mass-produced resume projects. #LLMs#Email#Automation

    InboxPilot AI is an AI-powered email management platform that uses NLP and LLMs to classify, prioritize, summarize, and analyze emails while generating context-aware replies. Built with React, TypeScr…

    Anirudh10-pngLanguage unknown★20

  • D coursera-assistant A Chrome extension that auto-plays Coursera videos, lectures, and discussions, plus AI-answered quizzes. It appeals to students wanting to save time, but it amounts to bypassing course requirements and fails on terms of service and academic integrity; technically it's just page script injection. #Browser Extensions#Automation#Learning

    Coursera Assistant, a Chrome extension for Coursera video completion, lecture completion, course automation, reading, discussions, Coach dialogues, and AI-assisted quizzes.

    Cicada33016HTML★8

  • D ps5-webkit-autoloader-13x The repo has no description at all; from the name one can only guess it's a WebKit autoload entry point for PS5 13.x firmware. Security researchers might be interested, but with no version, exploit chain, or usage, outsiders can't verify it or judge reliability. #Exploits#Reverse Engineering#Hardware

    CielWhatC★12

  • D CV_Deployment_skill Claimed to be computer vision deployment resources and engineering pipeline configuration, but the description is so vague you can't tell whether the deliverable is scripts, docs or templates. With no tech choices or examples, it's hard to judge which part of the deployment chain it addresses. #ML Research#DevOps#Agent Skills

    Computer vision deployment resources and engineering workflow configuration

    LMIXRLanguage unknown★10

  • D job-scraper No description; the name suggests a job scraping script. Which sites it scrapes, what format it outputs and how it handles anti-bot measures are all unknown, so job seekers or recruiting-data people get no usable information. Wait until the author adds target sites and usage. #Jobs#Automation#CLI

    divyesh123-jainPython★9

  • D the-arc A short film about war, labeled as model-generated. It can be viewed as a sample of AI-generated imagery, but it is a content work rather than a tool, and the generation process is not disclosed, so it has no reuse value for developers. #Video#Content Writing#Animation

    A video about how humanity went too far in war.Made By Opus5.5

    jasontzeng123TypeScript★14

  • D doom-os Claims to be the official DOOM OS, with an HTML tech stack, and does not explain whether it stuffs DOOM into the boot entry or simulates it in a webpage, nor does it give installation steps. It is more meme than engineering; unless a runnable image appears later, it is not worth opening. #Games#Operating Systems#doom

    The Offical DOOM OS repository

    klangfeld-labsHTML★16

  • D ios-messiah-re Empty description; the name suggests reverse engineering of some iOS target. It says neither what the target is nor what the analysis found or how to reproduce it, leaving readers with only a悬念. Reverse-engineering repos most need their motivation and boundaries stated, and this has neither. #iOS#Reverse Engineering#Debugging

    shiedlessLanguage unknown★22

  • D Hybrid-AI-Assistant Yet another "offline AI assistant for everyone," with only a slogan and no model choice, hardware requirements or install instructions. Mature offline assistants already exist, and this one offers neither differentiation nor evidence that it actually runs. #AI#LLMs#Speech

    This is an offline AI assistant for everyone

    siddikethanaHTML★9

  • D HouseOS Only one line, 'your home's operating system,' with no hardware target, no specific problem solved, and nothing installable in sight. Home automation is already occupied by Home Assistant, and 9 stars in a day proves nothing. Wait until it explains the first thing. #Operating Systems#IoT#Automation

    Your home's own operating system

    boitech-devPython★9

  • D python-primery-koda-2026 A Russian-language collection of Python 3.14, asyncio and FastAPI examples and exercises, belonging to tutorial-note content. It may be handy for Russian-speaking beginners, but as a repository it has no tool value and nothing new. #Learning#Documents#Content Writing

    Python code examples 2026: best practices, example walkthroughs, Python 3.14, asyncio, FastAPI, pytest, exercises with solutions in Russian

    justxorPython★9

  • D Real2Gym Nothing but a name: readers cannot tell whether this turns real scenarios into simulation environments or is a reinforcement learning training ground. A repo with no README information cannot be evaluated today, let alone used, so it is set aside for now. #Robotics#ML Research#3D

    real2gymPython★9

  • D htlb A mobile reading page and PDF for a "high-value life guide," this is pure content curation with no tool or method contribution. It may be a useful checklist for readers, but it is not an evaluable software project and does not deserve a ranking slot. #Knowledge Base#Documents#Health

    HowToLiveBetter — a cost-effective life guide: mobile reading page + PDF

    yuf22300-lmyfyHTML★9

  • D gridwatch-ai An 'explainable AI platform' for power load forecasting and anomaly detection that piles on LSTM, XGBoost, SHAP, and RAG, without saying where the data comes from or what the metrics are. There's no reproducible experiment for grid practitioners, and for developers it's just a list of technical terms. #ML Research#Data Analysis#AI

    GridWatch AI is an explainable AI platform for electricity-demand forecasting, anomaly detection, load-pattern clustering, and grid-risk assessment. It combines LSTM, XGBoost, Isolation Forest, K-Mean…

    Anirudh10-pngLanguage unknown★20

  • D HD2-Target-Overlay A mod that overlays objectives on the minimap in Helldivers 2, essentially a wallhack-style cheat in a multiplayer game. There's little technical explanation and obvious compliance risk; no one should install or spread it. #Game Cheats#Games#Reverse Engineering

    Helldivers 2 target minimap overlay mod, build scripts and research

    Samurai019Python★8

  • D wk-autoloader-relapse WK Autoloader on PS5, covering the umtx2 and relapse exploit ranges—a console jailbreak toolchain. Useful for players on specific firmware, but it involves cracking, carries high stability risk, and has little technical documentation. #Jailbreak#Exploits#ps5

    WK Autoloader for PS5 (umtx2 1–5.50, relapse 7.00–13.60). Current release v1.00.

    X-F1REBALL-XJavaScript★10

  • D Thinking_Reward_Model The repository has no description, and only the name hints at chain-of-thought reward modeling, with no data, method or reproducible experiments visible. People doing RLHF might be interested, but it is not even clear whether this is paper companion code, so there is no reason to open it. #ML Research#LLMs#Benchmarks

    bxhsortPython★16

  • D genpark-hubx-ai-apps-orchestration-skill It claims to distill hubx.co's product capabilities into Agent skills, including multi-model chat orchestration and visual diagnostics, but the description is all stacked product names with no interface, usage, or examples. There's nothing evaluable for users, and no visible difference from general agent frameworks. #AI Agents#Agent Skills#Multimodal

    GenPark AI Agent Skill - HubX intelligent AI apps portfolio, multi-model chatbot orchestration, vision diagnostics & generative tools distilled from hubx.co/products.

    Alpha-ParkPython★7

  • D zapret-discord The title is all Russian exclamation marks and 'new fixes', and the content points to configurations for bypassing Discord blocks. It explains neither the mechanism nor the applicable network environment, making it a typical traffic-bait repo; those who really need it will read the upstream zapret docs. #Networking#Network Security#Privacy

    NEW FIX How to Bypass the Discord Block in Russia?! New Discord Bypass for PC! Discord Ban & DS Fix!

    BeamBatAngularLanguage unknown★12

  • D ai-paper-system A one-stop service for ghostwriting papers, reducing AIGC detection and defense PPTs, plus open OEM recruitment. The tool's value is buried under marketing, and the use itself crosses academic integrity lines, so it's not worth developers' attention. #LLMs#Content Writing#Marketing

    AI paper system · AI paper generation, AI paper writing, one-click paper generation, AIGC reduction, plagiarism reduction, defense PPT · OEM white-label franchise recruitment

    FoShmilyLanguage unknown★29

  • D framework-kcm No description; a KDE-related component written in QML, with no way to tell its specific function or applicable scenarios. Without usage instructions it is impossible to judge whether it is usable, so it is not recommended for now. #Linux#UI Design#Hardware

    halfcyanQML★14

  • D csgo-julia-plugin No description; the name points to a Julia plugin for CS:GO, but it explains neither its purpose nor how to install it. It is impossible to tell whether it is data analysis or a game cheat, so per the rules it is not recommended. #Games#Plugins#csgo

    jrlucas1Python★22

  • D knowledge-agent Empty description; only the repo name hints at a knowledge-base agent. That space is already crowded, and with no word on how data gets in, how retrieval works, or who it is for, there is no way to tell it apart from existing RAG projects, so it is not worth opening. #RAG#AI Agents#Knowledge Base

    redwavetopPython★10

  • D relapse No text explaining what it does, just a name and a JavaScript implementation. Such a repo cannot be assessed for purpose or completeness, nor judged whether it is merely an empty shell of an idea, so it is not recommended. #Web Frontend#Productivity#AI

    soniciso1JavaScript★14

  • D psviethoa-webkit-autoloader The name points to some WebKit autoload or Vietnamese localization patch tool, but there is no documentation at all, and neither the install method nor the target is clear. Such language-pack/injection tools often skirt licensing gray areas, and with opaque information they are even less worth touching. #Plugins#Downloaders#Automation

    thanhsondevPython★19

  • D FilterTap Empty description; from the name and C++ implementation one can only guess it is some kind of filtering or audio-plugin signal routing, but there is no telling who it is for, how to integrate it, or whether it has been validated. A repo with no explanation cannot be evaluated no matter how new. #Audio#Plugins#Performance

    weak1337C++★20

  • D understanding-opd-crosscoders An empty-description machine learning research repo whose name points to some kind of cross-encoder interpretability analysis. With no abstract, experimental setup, or conclusions, it offers peers nothing readable and is currently just a placeholder directory. #ML Research#LLMs#Benchmarks

    yzc-666Python★10

  • D srl-lab-builder The README doesn't explain what it does, leaving only a repo name and an HTML page; it's unclear whether it's a lab builder or a course template, with no usage guidance. It's not usable by anyone yet, so wait for the author to add purpose and screenshots. #Web Frontend#Learning#UI Design

    0returnHTML★11

  • D BSB_sleepmod This reads like a sleep aid that reduces facial pressure with washable, replaceable printed parts, not a software project, and the repo contents can't be judged. Without drawings, materials or licensing, 3D printers can't reproduce it. #3D#Health#Hardware

    BSB sleep mod with reduced facial pressure, a washable interface, replaceable printed parts, and excellent ventilation.

    CucumberWorksLanguage unknown★8

  • D CVE-2026-20817 A CVE PoC repo with only numbers and no README, with the same author posting several in a short time. The vulnerabilities can't be confirmed to exist and there are no reproduction steps—low-quality placeholder, not to be used as threat intel. #Exploits#App Security#Reverse Engineering

    ZeroDayEvilLanguage unknown★10

  • D CVE-2026-41651 Like the author's other two repos, it has only numbers and no body text, so the vulnerability can't be verified. Security researchers need reproduction conditions and impact scope, and there's none of either here. #Exploits#App Security#Reverse Engineering

    ZeroDayEvilLanguage unknown★9

  • D CVE-2026-75604-PoC The title says PoC, but the repo has no description or code clues, and the numbers can't be checked either. This kind of bulk CVE placeholder repo is neither useful for defense reference nor worth bookmarking. #Exploits#App Security#Reverse Engineering

    ZeroDayEvilLanguage unknown★9

  • D Online-Voting-System-in-django A Django online voting system, a typical coursework project: registration/login, candidate list, vote counting. The description is long but has no security design, and 21 stars don't match the content. #Learning#Databases#Web Frontend

    An Online Voting System developed using Django provides a secure and user-friendly platform for conducting elections digitally. It allows users to register, log in, view candidates, and cast votes onl…

    anushap6611-sourceLanguage unknown★21

  • D openlawtw Empty description; the name suggests something to do with Taiwanese legal texts, but it is unclear whether it organizes statutes, searches them, or something else. Legal projects live or die on data sources and update methods, and none of that is explained, so usability cannot be judged. #Documents#Knowledge Base#Learning

    termcavetwHTML★19

  • D witcher3-crossover-fix Appears to be some compatibility fix patch for The Witcher 3, but not a word on which problem it fixes, which versions it applies to, or whether it conflicts with other mods. Players can only try it blindly, and it offers developers no reference value. #Games#Debugging#Windows

    tholtman1-delC★8

  • D deepseek-chan-emotes A pack of 161 DeepSeek-chan GIF stickers, purely an asset repo. It's mildly interesting for chat users who like the character, but it's neither a tool nor technically substantive, and its stars come mainly from character popularity rather than project value. #Images#Animation#AI

    Blue Big Fat Fish · DeepSeek-chan GIF emote pack | 161 GIF emotes exported with EmoteLab · 500px · 30fps

    BrickAZLanguage unknown★17

  • D react-fundamentals-lab-starter A starter template for the React exercise in section 8 of a course, left blank for students to implement themselves. Useful to instructors, worthless to anyone else, a typical course assignment repo. #Web Frontend#Learning#UI Design

    Projeto inicial da Aula 08 de SPA para reforçar fundamentos do React. Inclui estrutura base, componentes comentados, assets e CSS completo, sem as soluções implementadas. A atividade aborda JSX, props…

    EduardoVitorInocencioCSS★20

  • D flux-2-max-api-ja An API relay and resale service billed as 'one key for 307 models,' using a price list instead of technical documentation. Quota sources and compliance are unclear—a gray-market resale page, not recommended for integration. #Scams#LLMs#APIs

    AI API gateway / relay: 307 models behind one OpenAI-compatible key. image2.5 $0.0085/image, Seedance 2.0 Mini $0.01056/sec, LLM from $0.0228/M tokens. USD settlement, pay-as-you-go, $1 minimum top-up…

    apimarta7irvnLanguage unknown★8

  • D reverse-proxy-api-id One of a batch of API relay sites, claiming to index reverse-engineered and official routes together with discounts. Essentially reselling other vendors' model quotas with unverifiable sources; the risk outweighs the convenience. #Scams#LLMs#APIs

    Rute rekayasa balik dan resmi dalam satu indeks: ID `-ext` adalah label balik, sisanya rute resmi berdiskon. AI API gateway / relay: 307 models behind one OpenAI-compatible key. image2.5 $0.0085/image…

    apimartahfdrlmLanguage unknown★8

  • D Air-Quality-Index-prediction_with_deployment An entry-level ML project predicting air quality from PM2.5 and other weather data, with deployment scripts—reads like coursework. No data sources, metric comparisons, or live URL, so reference value is limited. #ML Research#Learning#Data Analysis

    Air Quality Index (AQI) Prediction with Deployment The Air Quality Index (AQI) Prediction with Deployment project is a machine learning-based application designed to predict air quality using environm…

    bhavyasreekovuri25Language unknown★9

  • D personal-resume-web-show A personal resume webpage with no description; the name suggests a practice or portfolio project. It explains no technical tradeoffs and extracts no reusable components, so it has almost no value for others and is not a candidate today. #Content Writing#Web Frontend#Jobs

    everynight31JavaScript★10

  • D Delta-Force-External Both the name and content point to an external cheat for a certain shooter game, with no description and no source explanation. Such repositories usually carry ban risks and executables of unknown origin, and do not meet this list's inclusion standards. #Game Cheats#Games#Windows

    exithostsparkLanguage unknown★12

  • D avcoe-code-roaster No description; the name suggests a tool that uses AI to roast code, but there is no implementation explanation or sample output. It cannot be evaluated and is not recommended. #Code Quality#LLMs#Testing

    omdesale777TypeScript★28

  • D IONForge_52 No description; neither the repository name nor the HTML tech stack shows what it does, and there are no usage instructions. It cannot be evaluated and, per the rules, is not counted as a recommendation. #Hardware#Embedded#Build

    garvitsahaHTML★18

  • D aegiscloud "Autonomous self-healing zero-trust cloud security + generative AI" packs four buzzwords into one sentence, with not a single executable detection or response logic in sight. Security products least tolerate empty vision; this description can neither be installed nor trusted, so skip it. #Network Security#AI#DevOps

    Autonomous Self-Healing Zero-Trust Cloud Security Using Generative AI

    siddikethanaLanguage unknown★8

  • D Deskrawl-modkit A 38-module cheat and trainer collection for Deskrawl, covering combat, economy, towns and more. A typical game-cheat repo with nothing to recommend it. #Game Cheats#Games

    38-module cheat/trainer for Deskrawl: Idle ARPG — combat, minions, economy, town and world tools

    WindoaretreatLanguage unknown★11

  • D Awesome-Physical-AI A paper list for embodied intelligence, a pure directory-style repo. There are already many similar awesome lists, and this one states no selection criteria and shows no sign of ongoing maintenance. #Learning#Robotics#AI

    Awesome Physical AI related papers

    HexaFutureLanguage unknown★9

  • D Discord-Nitro-Generator A so-called Nitro generator that is essentially a clickbait fake tool, and 167 stars in half a day is typical star-farming for traffic. It can't actually produce redemption codes and isn't worth anyone downloading or running. #Scams#Social#Automation

    OutsideglutweezersLanguage unknown★167

  • D InfiniHand No description; the name suggests something related to dexterous hands or robotic arms, but there is no hardware, algorithm or usage explanation. It cannot be evaluated and is not recommended. #Robotics#Hardware#IoT

    infinihandLanguage unknown★8

  • D OpenZilo The description is only the slogan "building the future of embodied AI," with no indication whether it is a framework, dataset, or hardware interface, and no getting-started path. Such vision-only repos cannot be evaluated and are not worth developers' time. #Robotics#AI#LLMs

    Build the future of Combodied AI

    ziloaiPython★12

  • D Flash-USDTBTC-Sender The name is the so-called lightning transfer tool, typical cryptocurrency scam language, with no description and no code explanation. Not included. #Scams#Crypto#Blockchain

    kerbastien5643Language unknown★21

  • D sidedoor No description; the name suggests some kind of backdoor or bypass channel, with no explanation of purpose or usage. Its nature cannot be judged, so it is not recommended. #Network Security#Privacy#CLI

    lassejlvRust★8

  • D audio-drama-release A release-only repo with no description or language, apparently distributing radio drama resources based on the name. It explains neither content source nor licensing and has no tool value, making it a typical asset repo that isn't an evaluable open-source project. #Audio#Streaming#Downloaders

    CopikProjeckIdLanguage unknown★16

  • D Forma A JS repo with not even one line of description, so it's impossible to tell whether it's a form library, a design draft or a half-finished product. Wait for the author to add documentation and usage. #Web Frontend#UI Design#Productivity

    GPTchatlyJavaScript★11

  • D Delta-Force-Cheat A cheat tool for Delta Force with not even a README. Neither compliant nor technically useful as reference, so it's an automatic reject under the rules. #Game Cheats#Games

    VisibleMaharajaLanguage unknown★12

  • D GXV-CISD-Model A repo with no description and no language, leaving only an abbreviation of a model name. It's unclear what problem it solves and there are no code clues, so it can't be evaluated. #ML Research#AI#Science

    Gregorius-XVLanguage unknown★11

  • D GXV-HTF-Profiler Another zero-description repo whose name points to some kind of "HTF" analysis tool, but with no input/output or usage described, readers have no way to start. #ML Research#Data Analysis#AI

    Gregorius-XVLanguage unknown★11

  • D gitContext The README is blank, and you can only guess from the repo name that it's related to git context; it doesn't even say what it does or how to install it. Projects with no stated purpose are out by the rules; wait for the author to add documentation. #Debugging#Code Quality#CLI

    PandyaPreetRust★12

  • D Caca-bug A Java repo with no description and no language, whose name suggests a classroom bug-hunting exercise. There's no explanation or runnable content, so even its purpose can't be determined. #Debugging#Code Quality#Learning

    FiamaBrendaJava★16

  • D entropy-vector A Python repo with no description and no language, whose name points to entropy vectors, but with no context or usage, it's impossible to judge whether it's worth opening. #ML Research#AI#Data Analysis

    LuxiaSLPython★12

Show HN 5

S

Show HN: Raven – The harness of harnesses, built for RSI

Show HN

Raven calls itself a 'harness of harnesses,' building a self-evolving multi-agent ecosystem to support RSI (recursive self-improvement), with 4,807 stars and 53 HN points—one of the strongest early momentum in this batch. If the RSI narrative pans out, it could become the infrastructure layer for agent orchestration. The risk is that the concept is ahead of its time, engineering feasibility is doubtful, and it competes head-on with mature orchestration frameworks like LangGraph.

#AI Agents#AI#ML Research

▲53

B

Show HN: NSL – WSL for Linux

Show HN

Uses a single VM plus systemd-nspawn to replicate a WSL-style multi-distro development experience, solving the pain of atomic-distro users repeatedly reinstalling environments—a real developer workflow improvement. 74 HN points acknowledge the need. It could become a standard Linux dev environment tool, but differentiation and maintenance sustainability are risks against mature options like Toolbox and Distrobox.

One of the things that Windows really got right is WSL2. I drive an atomic Linux distro for daily use, but wanted a way to develop with multiple different distros with that same WSL UX. NSL is my answ…

#Linux#DevOps#CLI

▲74

B

Show HN: Jevstiller – Distill Jev into a local model, with a disagreement bound

Show HN

Jevstiller proposes distilling Jev into local models with a 'divergence upper bound' guarantee, adding verifiability to lightweight decision models—a rare piece of rigor in the Jev ecosystem. 59 HN points show community interest. But the theoretical guarantee is far from production-ready, and the whole Jev paradigm is unvalidated by the market, with an unclear commercialization path.

#ML Research#AI#LLMs

▲59

B

Show HN: Real-time Solar System with 526k asteroids and all tracked satellites

Show HN

Turns 526,000 asteroids plus all orbiting satellites into a real-time 3D visualization, with the data pipeline and rendering scale showing real engineering capability. 81 HN points prove its appeal for public and educational scenarios. Commercialization lies in space education and B2B satellite operations visualization, but such one-off demos rarely grow into standard dependencies; it depends on sustained maintenance and data updates.

#3D#Data Analysis#Science

▲81

B

Show HN: Pac-Bench – How well can models one-shot a Pac-Man game?

Show HN

Pac-Bench evaluates models head-to-head with a single prompt to generate a Pac-Man game, a lightweight benchmark of harness plus model capability. Its HN score of 78 shows interest in one-shot generation quality, and it could gain momentum as a cited community benchmark, but the task is too narrow and easily replaced by more comprehensive coding benchmarks, limiting commercial value.

Benchmarks how well Harness+models can create a Pac-Man game from a single prompt: “Create a Pac-Man game in a single HTML page” Each model gets one shot — no follow-up prompts or fixes.

#Benchmarks#Games#AI

▲78

Hot New Repos

  1. KKKKhazix/AIHOT★3087 A website framework that finds trending topics on its own and writes its own daily digest. Swap in your own sources and curation criteria, and it becomes your industry trending site.
  2. firelex/jeff★1011 Fine-tunes of Qwen3.5 and Gemma 4 for zero-shot classification
  3. wy51ai/floorplan-3d★478
  4. PostHog/jeeves★267 Jeeves – Reasoning improves Jev-like decision models
  5. openJiuwen-ai/iCode★238 A lightweight, extensible, fully offline development platform and agent/workflow toolkit for the AI era, with an intuitive TUI, easy deployment, and complete control over your data, agents, and workflows.
  6. openai/mcp-extensions★208 Build plugins that feel like native, first-class features of ChatGPT.
  7. BagasRizkyHarySaputra/MLBB-waydroid-LinuxCloudMLBB★196 Play Mobile Legends: Bang Bang (MLBB) on Linux via Waydroid — and stream it to your Android phone as cloud gaming (Sunshine + Artemis/Moonlight). Working multi-touch, Intel VAAPI encoding, one-command launcher.
  8. Outsideglutweezers/Discord-Nitro-Generator★166
  9. TreapGoGo/deepseek-whale-girl★137 A community-maintained fan resource library for the whale girl, organizing character reference images, generation prompts, community works, and related projects to provide creators with convenient references and inspiration. This is an unofficial community project and does not represent the official position of DeepSeek.
  10. ios-bundles/CarrierSIM★112

Star gain 1d

  1. debpalash/VoiceStudio+4951★46491
  2. eternity4719/HowToLiveBetter+3571★28081
  3. yetone/magpie+1052★2639
  4. cdyforever/how-to-live-better+1032★4890
  5. NandhaKishorM/laya+914★28600
  6. firelex/jeff+789★1011
  7. affaan-m/ECC+716★268030
  8. Niko1221/Strata+655★1717
  9. DietrichGebert/ponytail+614★147093
  10. Human-Agent-Society/reef+444★7321