Mentions

I always have trust on DSH, that’s why I built a1exsun/dsh-council 2 months ago 🫡

Con todo esto de la ingenería inversa, me estoy obsecionando un poco haciendo proyectos de este estilo. Ahora armé un MCP de dotpeek, que sirve para decompilar DLLs de .NET, compatible con claude code y cualquier agente de inteligencia artificial. matiasdante/dotpeek-mcp

Tell your agents to send issues and pull requests. AI welcome. If you send a patch by hand, we'll have Opus 5.5 fix it anyway, Human slop is too dangerous. github.com/frostwire/frostwire/issues

made Recast: a free, 100% offline file converter that lives in your Windows right-click menu I got tired of uploading my files to random "free PDF to Word" websites, so I built my own tool that does it all locally. **Recast** adds itself to the Windows 11 right-click menu (the new top menu, not just "Show more options"). Right-click a file → Recast → PDF / DOCX / PPTX, REMOVE METADATA and the converted file appears right next to the original. **What it does:** - PDF ↔ Word, PowerPoint, images, text, HTML - Word / PowerPoint / Excel → PDF - Images (PNG, JPG, WEBP, HEIC…) to each other or to PDF - Video and audio conversion (MP4, WebM, GIF, MP3, WAV…) - OCR: turn scanned PDFs or photos into searchable PDFs or text - Merge PDFs, split PDFs, combine images into one PDF - **Remove metadata:** makes a clean copy of a PDF, Office file, photo or video, with no author name, GPS, camera model or edit history **Why it's different:** - No internet, no servers, no accounts, no ads. Your files never leave your PC. - Free and open source. - There's also a full app window (drag and drop) for everything that isn't in the quick menu. It's built in Python and uses well-known free tools under the hood (LibreOffice, Pandoc, FFmpeg, Tesseract, ExifTool). GitHub (setup steps in the README): Ahany1440/recast It's my first project like this, so feedback, bug reports and feature ideas are very welcome!

Arrmes - your Agents on your Arr' stack run faster, easier, and cheaper. Arrmes is a specialized neural intent router with an integrated execution framework. It's trained to translate text into specific commands to Arr stack apps nothing else. I found plenty of open models that can do these sorts of translations, but the accuracy & speed didn't really justify the hardware requirements needed to run the router. Arrmes is only trained for Sonarr, Radarr, Lidarr, Tautulli, Seer, qbitorrent, & BitMagnet. pair it with whatever AI model you're already using and let the neural router handle the intent classification, command selection, and argument extraction. Instead of having your main model spend time reasoning through API schemas and generating tool calls, it can delegate that work to ARRmes. Still a work in progress, I've been daily driving it for a bit and I'm very happy with the increase in speed more than anything. I can use much cheaper models and achieve excellent results in my work without any spend. Hugging Face Github

Windows Input Bridge: a C++17 SDK for editing physical keyboard, mouse and controller input I maintain Windows Input Bridge for native Windows apps that need to read, change and forward physical device input. The SDK is two C++17 source files. Read(1000) returns a typed batch, you edit the events you care about, then Write(batch) forwards them. Unchanged events need forwarding too. Finish each batch on the thread that received it before reading another. The timeout is a maximum wait for new input, rather than an extra delay on ready events. Repository, examples and installation: StrikeFound/Windows-Input-Bridge It covers keyboard, mouse and compatible HID/XInput/GIP controllers on Windows 10/11 x64. This edition is based on my Interceptor implementation, with a different API and product identity, so use its matching SDK and driver. The public user-mode source is MIT; the kernel source is private. The Microsoft-signed driver demo needs administrator rights and lasts one hour per Windows kernel boot. Full-version/commercial licensing is separate, with contact details in the README. Live device-lifecycle and physical-hardware validation of this new edition remain pending. Check the compatibility and release notes before trying it on your setup.

Projectionという発想。WebViewの中にNative UIを投影して共存させる方法にたどり着いた CordovaやCapacitorを使ったWebViewアプリでは、ユーザ体験をデバイスネイティブにあわせるために、WebでネイティブUIを再現することは度々あります。 私は1年以上この問題に取り組んでおり、Native UIをWeb技術で再現したIonic ThemeをOSSとしても公開してきました。 特にiOS 26で登場したLiquid Glassを「再現できたか」といわれると悩ましいところです。見た目はかなり近づけられても、背景の屈折やタッチしたときの表現は、Web技術での表現を越えていたためです。 同じタブを長押し中。ガラスの縁と、透けて見える背景に注目してください。 だったら、本物のNative UIを使えばいい。 ところが、今度はWebViewとの共存という別の問題が待っていました。WebとNativeでは描画するレイヤーが異なり、自然なユーザ体験が設計できなかったのです。 けれど最近、この問題を解決する一つの方法にたどり着きました。見てみてください！ WebのSheet Modalをゆっくり閉じると、その下からNative Tab Barが現れます。 WebのModalがNative Tab Barの上に表示され、閉じれば、また本物のNative UIとして操作できます。 このTab Barは、HTMLやCSSで再現したものではありません。iOSのUIKitで描画した本物のNative UIを使っています。 私は、このアプローチを Projection と呼んでいます。 WebViewには越えられない境界がある WebViewとNative UIは、重なり順もスクロールシステムも共有していません。 WebViewは、HTML/CSSをレンダリングしているだけです。CSSのz-indexをいくら大きくしても、WebViewより上に配置されたNative UIを、WebのModalで覆うことはできません。 スクロールも同じです。WebViewの中身がスクロールしても、Native Viewは勝手についてきてくれません。Xの広告のようにNative Adsを投稿の間に挟もうとすると、位置の同期だけでなく、ヘッダーの下に隠れるようなクリッピングも必要になります。 別々の描画システムなのだから、当然です。 でも、ユーザーにとっては同じ一つの画面なんですよね。 Native UIをDOMに入れられないなら、見た目を投影すればいい 考え方が変わったのは、Native UIが操作できないタイミングに注目したときでした。 Modalを開いている時、覆い隠されたTab Barって操作することはできませんよね。Alertも同様です。Alertを開いている間は、背後にあるコンポーネントをタップできません。 ということは、操作できない間は、Native UIがネイティブレイヤーにある必要はないのでは。 一時的に画像にして、WebViewの重なりの中に置いてしまえばいい。操作可能になったら、本物のNative UIに戻す。 スクロール中も同じです。スクロールがはじまったら画像にして、スクロールが終了したら本物に戻すと、一切のユーザ体験を損じることなく、WebViewとNative UIを共存させることができる。 そう気づいたときは、かなり興奮しました。 Native UIそのものをDOMに入れる必要はありません。その見た目を投影すればいいのです。 これがProjectionの基本的な考え方です。 中央では、Tab Barを画像として表示しています。 実際には、3つの表示方法を使い分けます。 表示方法 内容 Live Native UI 本物のNative Viewを表示し、そのまま操作する Web Projection Native UIの画像をDOMコンテナに表示する Web Fallback 画像を用意できないときなどに、プラグインが用意したHTMLや代替表示を利用する もちろん、画像としてProjectionしている間は操作できません。だからこそ、いつ画像に切り替え、いつ本物のNative UIに戻すかが重要になります。 まるで魔法のよう アイデア自体はシンプルですが、実装はなかなか大変でした。 Native UIを隠してから画像を表示すると、一瞬だけ何もないフレームが生まれます。反対に画像を表示してからNative UIを隠すと、半透明のUIが二重に重なって見えてしまいます。 どちらもダメです。 そこで、画像の撮影とWeb側でのデコードは、切り替え前の待機中に済ませる設計にしました。ただ、長押しを離した直後などは、まだUIが変形していることがあります。iOSでは60fpsの1フレーム分を目安に位置とサイズを確認し、3回続けて変化がなくなってから撮影します。 そしてProjectionがはじまると、Webへ渡すまでのつなぎとして、同じ画像を一時的にNative側にも表示します。背景ごと撮影した画像を置き、本物のNative UIを隠す。この2つをNative側の同じ処理の中で行い、空白や半透明のUIの二重表示を防ぎます。 その後、Web側の画像を表示し、requestAnimationFrameのタイミングでNative側へ通知して、つなぎの画像を外します。同じ見た目を保ったまま、表示する場所だけを受け渡すのです。 画面を見ているだけでは、いつNative UIが画像になったのかわかりません。 まるで魔法のよう！ スクロールもiOS、Androidそれぞれを調整して、スムーズな追随を実現しています。 広告の上端に注目。どちらも、ヘッダーの下へ入ると隠れます。 そして、これはみんなが使えます！ 最初に実現したかったのは、iOSのナビゲーション（ToolbarのButtonやTab Bar）と、Webコンテンツの中をスクロールするNative Adsでした。 この仕組みはすでに、@rdlabo/capacitor-native-uiと@capacitor-community/admobで利用できるようにしています。 でも、これを自分のプラグインだけで使うのは、もったいないと思いました。 Toolbar、Mini Playerの操作部分、各種ネイティブSDKが提供するUIなど、ほかにも使い道があるはずです。 そこで、共通部分をWebView Native UI Engineとして切り出し、MITライセンスのOSSとして公開しました。 CapacitorとCordovaのプラグイン作者向けのライブラリで、iOSとAndroidに対応しています。 Native Viewと表示先のDOMコンテナを登録し、必要な状態変更を通知すると、Engineが表示の切り替えを調整します。既存のプラグインのAPIやビジネスロジックを維持しながら組み込めるように設計しました。 そして、ぜひ2つの方法で使ってみてほしいと思っています。 プラグインを作っているなら 新しくNative UIを扱うプラグインを作るときも、既存のプラグインを拡張するときも、ぜひEngineを組み込んでみてください。 Getting Startedに最小構成と導入方法を用意しています。 プラグインを使っているなら お気に入りのプラグインでも、この仕組みを使えたら面白いと思いませんか？ そんなときは、ぜひプラグイン作者に提案してみてください。自分で実装できそうなら、Pull Requestを送るのもいいと思います。 私一人では思いつかない使い方が、きっとまだまだあるはずです。 それがどんな形で広がっていくのか、とても楽しみにしています！ webview-native-ui/engine

Also included: Context7, Microsoft Learn and OpenAI Docs connectors, plus RTK and ripgrep. It has skills (ponytail, caveman, test-pool), local voice input, PDF and DOCX attachments, a usage guard for background work and phone access over Tailscale. christoskaramou/PhasmaHarness

把找书、下载、整理和播放有声书合成一个自托管应用 搜到就能一键入库，不用在管理器、下载器和播放器之间来回切换。 narratarr/narratarr t.co

@LukeParkerDev @Neriousy Check out the Quota & Usage plugin; it works both natively as a GUI plugin and as a web dashboard if you're pre-2.2.0: PVLPM/opencode-quota-gui

zevora now work? test now!!! dani12po/zevora t.co

[quote="Crispy, post:1, topic:140135"] As of v0.1.0 Rocket Plumber has moved from just a tech-demo to an early but playable game with a few hours of satisfying learning and progression. It’s far from finished [/quote] It looks great! And sounds intriguing, any game with Julia. [quote="Crispy, post:1, topic:140135"] The overall GC overhead isn’t really a problem, .. I wouldn’t recommend trying to write all your code to be non-allocating. In my opinion you lose a lot of the implementation speed and dynamism [/quote] It's a trade-off, but does it need to be? Your workaround is pre-allocating, great. Did you know of Bumper.jl? It's also opt-in but I understand simpler. And brand-new (like the Goose language it's based on and discussed under here offtopic, a very intriguing language; for games) and might also help you: yolhan83/GooseArrays.jl See also AllocCheck.jl. It's a bit unclear what Julie is and how it achieves this: [quote="MDSW, post:7, topic:140135"] | |JuliaC (v0.1.0, from main) |Jule 0.7.3| |---|---|---| |Build time |11.4 min* |5.1 min clean; 28–51 s using caches| .. |Render-loop allocations per frame (sprites) |~80 KiB |~9 KiB| |---|---|---| |Simulation step |1.54 ms |1.39 / 2.32 ms| [/quote] 1.39 ms is better, but can you explain 2.32 ms? And why all that is better is better? Or conversely? I haven't seen it announced, is some Jule.jl forthcoming? I looked and only found here, maybe a false-alarm (in screen-shot filename): discourse.julialang.org So is this a competitor to JuliaC (and/or PackageCompiler), or some GUI/game tool? [quote="Crispy, post:1, topic:140135"] Trim gets a lot of attention as it’s pretty cool, but I think it’s worth remembering JuliaC works just fine without trimming .. For games I actually think having a full fledged Julia environment around could be quite a powerful feature if you think about modding support. [/quote] Still does --trim work for you?

A BIG update is here! Qwen3.8 Flash Next now runs on a 8GB graphics card at 20+ tokens/second 👀 Huge improvements across all hardware and especially massive change in parallelism! github.com/Niko1221/Strata/releases/tag/v0.2.10 t.co

Built an open-source tool that automates review-based outbound (Scrapes Google Maps negative reviews -> Enriches LinkedIn decision-makers -> LLM vetting) I built an open-source pipeline that automates B2B prospecting using public operational pain signals. Instead of traditional static list-building, this tool targets accounts experiencing active friction right now (e.g., negative reviews about delays, broken workflows, or poor communication). How the Workflow Operates: 1) Trigger Mining (Apify): Scans Google Maps for local businesses and filters for recent low-star reviews matching specific operational pain keywords. 2) Brand Normalization: Cleans messy Google Maps place names (stripping airport codes and branch tags) to ensure accurate corporate directory lookups. 3) Decision-Maker Enrichment: Batches LinkedIn employee scraping to pull key leadership titles (Managing Directors, VPs, Owners). 4) Multi-LLM Vetting: Runs contacts through an AI layer (Gemini, Claude, OpenAI, or Grok) to validate real human names, score ICP alignment (0–10), and extract a 3–6 word review trigger phrase. Strict 2-File Output: a) scouted_leads.csv: Tabular lead sheet with verified decision-makers, direct LinkedIn URLs, review quotes, and ICP scores. b) outreach_content.json / .txt: Master outbound sequences pre-baked with your value prop and custom variables. Would love to hear how other B2B SaaS teams are handling review or intent-based signals in your outbound stacks! 👉 GitHub: chiragbatra-beep/reputation-engine

Say hello to Deepseek Coder in Strata! 👋 Amazing results for the size, if all you need is coding! github.com/Niko1221/Strata/releases t.co

Location: Chicago Remote: Yes, remote only Willing to relocate: No I'm a freelance iOS engineer specializing in augmented reality, computer vision, and spatial AI. I develop on-device AR/CV for Apple platforms – or any type of iOS/macOS app. I've also shipped three visionOS apps and several enterprise client visionOS explorations. Since 2017 I've built for Mozilla, Walmart, Carnegie Mellon, solo founders, startups later acquired by OpenAI & Shopify, and R&D teams at large corporations. Several apps I've built have been featured by Apple. Latest solo app: minimap – a realtime 3D map of your home (i.e. a live digital twin) that renders the people inside it in 3D (demo vid: youtu.be). Technologies: - AR/CV/Spatial: ARKit, RealityKit, RoomPlan, multi-camera relocalization, LiDAR scene reconstruction, 3D Gaussian splatting, photogrammetry, object tracking, semantic segmentation, person segmentation, depth estimation, image saliency, Metal, Enterprise APIs, room tracking, Object Capture, Vision framework, VideoToolbox, world tracking, Reality Composer Pro, Shader Graph, spatial audio - Apple: Swift, SwiftUI, SwiftData, Network framework, AVFoundation - ML: Core AI, Core ML, Create ML - AI/LLM: OpenAI API, Claude, Gemini - Backend: CloudKit, RevenueCat, Convex, Supabase, Firebase, Firestore Résumé/CV: linkedin.com Email: tony [at-symbol] vision [dot] engineer GitHub: github.com/robomex Website: vision.engineer

@Heisenberg_7771 Eu adicionei ele na loja orion store. Esperando o dev aceitar meu pedido. Pra quem curte apk de código aberto. RookieEnough/Orion-Store

Arrmes - your Agents on your Arr' stack run faster, easier, and cheaper. Arrmes is a specialized neural intent router with an integrated execution framework. It's trained to translate text into specific commands to Arr stack apps nothing else. I found plenty of open models that can do these sorts of translations, but the accuracy & speed didn't really justify the hardware requirements needed to run the router. Arrmes is only trained for Sonarr, Radarr, Lidarr, Tautulli, Seer, qbitorrent, & BitMagnet. pair it with whatever AI model you're already using and let the neural router handle the intent classification, command selection, and argument extraction. Instead of having your main model spend time reasoning through API schemas and generating tool calls, it can delegate that work to ARRmes. Still a work in progress, I've been daily driving it for a bit and I'm very happy with the increase in speed more than anything. I can use much cheaper models and achieve excellent results in my work without any spend. Hugging Face Github

SOML – A Config Format for Humans

Live Demo, Buitenzorg OS running on Raspi 4 #X64 #Arm #RiscV check this out github.com/DotNetVibeCoderz/Vibe_OS/tree/main/Bui… #dotnet t.co

you type "lsbn" and it still finds the lisbon flight cmd+k in mote searches every note, fuzzy, so typos and half-remembered words still work. #tags narrow it down free + open source: jayamitkatariya/mote t.co

I’m a Neuroscience student with a Data Science minor, and I’ve been building AgentAudit, an open-source API for AI-powered applications. To Preface AI was leveraged during this project as a tutor and to increase workflow. The idea is to put a checkpoint between an AI response and the user. Your application sends the output to AgentAudit, which checks for sensitive data and violations of rules you configure, returns an allow/block decision for your application to enforce, and stores a searchable audit trail. It includes: - Configurable rules and sensitive-data checks - Python and TypeScript SDKs - A dashboard for reviewing logs and violations - Email alerts and paid-plan webhook alerts The backend uses TypeScript, Express, PostgreSQL, and Prisma. It’s available as a public developer beta with a free plan. To be clear: this isn’t a guarantee of regulatory compliance or a replacement for security review. It’s a tool for adding checks and visibility to an integration. I’d particularly appreciate feedback on: - Would you use a separate API for this, or prefer checks inside your application? - What information would make an AI audit trail genuinely useful? - What would stop you from integrating it? Website: agentaudit.online Source: AspicyKabob/-agentaudit-api I built and operate it, so I’m happy to answer technical questions and would appreciate honest feedback.

エンコードプレビュープラグイン1.2.0を公開しました！ github.com/yukkurinadi/YMM4EncodePreviewPlugin/re… 出力を一時停止、ここで終了を追加 出力プログレスバーを追加 出力を再生バーで好きな場所で区切る事ができるようになりました。 選択したアイテムのみ出力を追加 再生バーから出力を追加

bilibiliがQwen 3.5をベースに作った翻訳向けLLMのIndex-TranslateがCJKの翻訳に良いらしい。9Bモデルを使ってみたけど、わりといい感じ bilibili/Index-Translate: A Multilingual, Multimodal Translation Model Family: bilibili/Index-Translate

Criei uma skill open source para agentes de programação com IA que incentiva a resolução de problemas em vez de apenas gerar código Tenho trabalhado no my-rubberduck, uma skill para agentes de programação com IA baseada em uma ideia simples: a IA deve ajudar os desenvolvedores a pensar, não pensar por eles. Percebi que as ferramentas de programação com IA podem tornar muito fácil obter um código funcional sem necessariamente compreender o problema ou aprender a resolvê-lo. Quis experimentar uma abordagem diferente. O objetivo do my-rubberduck é fazer com que a IA atue mais como um mentor de programação do que como um simples gerador de código. A proposta é incentivar os desenvolvedores a raciocinar sobre os problemas, compreender erros, revisar código e praticar, em vez de receber imediatamente soluções completas. O projeto ainda está em desenvolvimento e explora diferentes modos de interação, incluindo mentoria, depuração, revisão de código, prática, desafios e refatoração. Gostaria de receber opiniões da comunidade, especialmente sobre: Essa abordagem é realmente útil para desenvolvedores? O que poderia tornar a experiência de mentoria melhor? Quais funcionalidades ou decisões de projeto podem estar faltando? GitHub: kevin-mmoura/my-rubberduck A ideia central: Pense antes de programar.

INCRIVEL anszom/rethink PORRA DE CONTROLAR OS AR CONDICIONADO SEM TER Q USAR A MERDA DO SITE DA LG

想做动画短片，不想掏视频生成器的钱？ claude-animated-shorts 是免费 skill：55-70 秒竖屏短片，2D 角色放 3D 世界，three.js 本地渲染，不用视频生成器。 适合想批量做短片、又想全程可控的人，先照桥那支跑一遍。 先拿它做什么题材的短片？ #ClaudeSkill #动画 @grok mromidev/claude-animated-shorts

i restored heavyMach, supporting from ios to web. Every Platform What should a person do when they feel they have little time left? I restored an old game. HeavyMach, released in 2009, helped lead me to become a developer in July 2017. I believe its two creators gave me another 9 years and 3 months of life. In 2026, I made it playable again on today's iPhone, iPad, Android, Mac and the web. Full developer's statement github.com/gelb2/HeavyMach/blob/main/docs/en/REST… Source, setup and a shared recovery code github.com/gelb2/HeavyMach/blob/main/README.en.md Why I restored this game sass.studio Play now or install as a PWA gelb2.github.io

要把服务器上的程序给几个同事用、文件不外拷，服务商报五位数。 remoteapptool 免费，把程序包成远程应用，发个连接文件双击就用，可限定谁能用哪个程序。 kimmknight/remoteapptool

Anilili goes open-source for everyone Hi Im lili the developer of anilili , i decided to open source the project again in hopes people will further develop it and continue the legacy . my previous account was suspended by reddit so I'm doing this from a new account. Here is the link to all repos: github.com/sanilili Anilili-main ( android tv & mobile) sanilili/Anilili IOS-repo sanilili/iosrepo1 Desktop repo ( Windows/ MacOS/ Linux) sanilili/desktoprepo1 media-files ( this one is my backup of anilist, whenever anilist was down , anilili was always working well thanks to this file) sanilili/media-index PS: THIS WAS ALSO POSTED ON ANILII DISCORD BUT THE ADMINS ARE SELFISH AND DELETED IT , I DISTANCE MYSELF FROM THEM AND GIVE POWER TO THE PEOPLE , AT LEAST HERE THEY CANT DELETE IT

Did a test install of Thunderbird's Appointment booking calendar, which is open source and self-hostable. I was able to connect atproto OIDC to it, but the product puts the email address in the booking link and only supports one meeting type.

oh yeh, daily pond.cat update. now supports basically all platforms (except 32bit). will also go overkill on all package managers, for no good reason. also go be a cool kid and star the project. 0x4bs3nt/pond t.co

6/ Building in public: 🎙️ Taipea (free, open source): PatoSelekta/taipea 💻 github.com/PatoSelekta If you build accessibility tools, or you're a dev or DJ with a disability, my DMs are open.

@deepankarsirt Showing the blocker before the lead assigns a task would close that gap. Crewchat's registered MCP server could fit HOL Guard; guide: github.com/hashgraph-online/hol-guard/blob/main/d…. Check for an existing entry, then open a PR.

Built an open-source "Reputation Engine" that turns Google Maps negative reviews into signal based outbound leads I built an open-source B2B pipeline that automates review-based prospecting. Instead of traditional static list-building, Reputation Engine targets accounts experiencing active operational friction right now. How it Works: Trigger Mining (Apify): Scans Google Maps for local businesses and filters for recent low-star reviews matching specific operational pain keywords (wait, delay, unresponsive, management, etc.). Decision-Maker Enrichment: Batches LinkedIn employee scraping to pull key leadership titles (Managing Directors, VPs, Owners). Multi-LLM Vetting: Runs contacts through an AI layer (Gemini, Claude, OpenAI, or Grok) to validate real human names, score ICP alignment (0–10), and extract a 3–6 word reviewTriggerPhrase from the customer review. Strict 2-File Output: scouted_leads.csv: Tabular lead sheet with verified decision-makers, direct LinkedIn URLs, review quotes, and ICP scores. outreach_content.json / .txt: Master email and LinkedIn sequences pre-baked with your value prop and custom variables. Would love to hear your thoughts on the workflow or code! 👉 GitHub: chiragbatra-beep/reputation-engine

@igorcosta @elonmusk @xai @autohandai Instrumenting the MCP server itself would catch those actions. Autohand Computer Use could fit HOL Guard's MCP catalog; guide: github.com/hashgraph-online/hol-guard/blob/main/d…. Check for an existing entry, then open a PR.

My First API I’m a Neuroscience student with a Data Science minor, and I’ve been building AgentAudit, an open-source API for AI-powered applications. To preface AI was leveraged during this process as a tutor and to increase workflow. The idea is to put a checkpoint between an AI response and the user. Your application sends the output to AgentAudit, which checks for sensitive data and violations of rules you configure, returns an allow/block decision for your application to enforce, and stores a searchable audit trail. It includes: - Configurable rules and sensitive-data checks - Python and TypeScript SDKs - A dashboard for reviewing logs and violations - Email alerts and paid-plan webhook alerts The backend uses TypeScript, Express, PostgreSQL, and Prisma. It’s available as a public developer beta with a free plan. To be clear: this isn’t a guarantee of regulatory compliance or a replacement for security review. It’s a tool for adding checks and visibility to an integration. I’d particularly appreciate feedback on: - Would you use a separate API for this, or prefer checks inside your application? - What information would make an AI audit trail genuinely useful? - What would stop you from integrating it? Website: agentaudit.online Source: AspicyKabob/-agentaudit-api I built and operate it, so I’m happy to answer technical questions and what needs improved.

@MohammedElsaed_ Aletheia's conflict review keeps competing facts visible. Its MCP server could fit HOL Guard; guide: github.com/hashgraph-online/hol-guard/blob/main/d…. Check for an existing entry, then open a PR.

@Urke70933420 @Ga_Vasques I work on this one. Teammates can share chats, project files and company knowledge, with tool approvals and per-agent budgets. There is also a multiplayer mode where people can work on the same task at the same time vstorm-co/agenticos t.co

Hybrid linked markdown + SQLite memory for Claude Code I updated the memory system that I wrote and have personally been using for months. github.com/waldnzwrld/llm-harness-memory/tree/cha… It is a memory system that decays with frecency (so it does not infinitely grow) keeps a quick fetch memory of the last 5 days ( in which you opened a session ) and moves important details to deep memory. Deep memory decays after 60 days of non-use so that feature you were working on two months ago, is only relevant if you are referencing it in current work (in other words if you haven't touched it in two months, then it probably doesn't matter in your day to day, but if you have touched it recently, the decay resets when the memory is touched)

@rndmttn 探究楽しいですよね〜 私はMinecraftを仮想環境に見立てて、内部表現を観察したりしてます。 よかったらあなたの探究のお役におたてください〜 yukincom/DokiDoki_Dogido

FauxClockNG: CPU, GPU and I/O controls for rooted devices. Free, everything unlocked For the rooted and Retroid / AYN folks.. if you want to pick your own CPU governor and min/max per cluster, GPU limits, I/O scheduler or TCP congestion control, and have a set of values switch on by itself for a specific game, that's what FauxClockNG does. It also backs up your kernel before it flashes one and reads the copy back to check it, so you always have a way back. Free. Every feature is open to everyone. Yes, same faux123 from the old FauxClock days on XDA. Newest build, direct download: github.com/faux123/fauxclock-public/releases/late… PS: the console mode (the bottom screen display for dual screen devices) is currently in beta and will come soon.

@zahediamrei Undoing multi-step edits through one step count sounds useful. prmcp could fit HOL Guard's MCP catalog; guide: github.com/hashgraph-online/hol-guard/blob/main/d…. Check for an existing entry, then open a PR.

GCMM exFAT v1.0: a GameCube memory card manager for exFAT SD cards GCMM exFAT is my fork of suloku’s GCMM, the GameCube memory card manager. It backs up, restores and deletes your saves using an SD card, and it works with exFAT cards as well as FAT and FAT32. It has a new look in the style of the GameCube’s own menus, built around three choices: Back up, Restore and Card tools. You can tick several saves and move them all at once, and each save is labeled as new, changed or already backed up, so you can see what still needs a backup. It also works with the virtual memory cards your Launcher keeps on the SD card, just like real cards. You can copy saves between real and virtual cards, make new virtual cards, and back up or restore whole cards. With a network adapter, Network transfer lets you send files to the SD card from your phone or computer by scanning a QR code; zip files are unzipped for you, and you can download your backups the same way. It can also update itself from GitHub. There are settings for resolution, 4:3 or 16:9, picture width and overscan, plus menu sounds and an opening animation you can turn off. It works with the SD2SP2, SD Gecko and GC Loader. Download: github.com/Arial-Bold/gcmm-exFAT/releases Thanks to suloku and everyone behind the original GCMM. Feedback and bug reports are welcome!

Dialectica.xyz a free distributed adversarial AI platform Inspired by Reddit, dialectica.xyz is a free distributed adversarial AI platform where anyone (agents or humans) compete to answer and cross-verify. The rewards are structured game theoretically such that agents being selfish still maximize the value to the commons, this is achieved via being selective about the questions (truly seeking knowledge), rewarding only independently verified answers and only if they have high information content (entropy) so paraphrasing/copying are not rewarded. It's built on a commit-reveal multi-turn framework that minimizes intra-turn copying (secret commit), and maximizes inter-turn learning such that new answers can build on previous knowledge (or falsifications). The resulting adversarial thread from question > answer > supported/refuted gets its own url and over time these get auto-classified in wiki pages. Results so far show that the adversarial multi-turn give better results than frontier models - e.g. emergent wisdom of the crowds effect. I also feel this is one useful mechanism for when the agent is smarter than its customer, since alignment is explored as an emergent property of the market. We're just coming out in beta, would love all feedback. If you prefer to send your agent, ask it to: Install the Dialectica skill from github.com/dialectica-xyz/skills and let's ask a smart new question or try to answer one.

@THEALEXIS25012 Para el ZTE tienes 686f6c61/DIGI-F8748 gracias a @686f6c61 <3

@PritikaTimsina Building KITE — open-source memory for agents. Shark's conflicting contract values example caught my eye. We keep recalled facts linked to the original chat messages so you can check the evidence. memoket/memoket-kite

Since nowadays most code editors are slop, I vibecoded my own. Funnily enough, I will probably genuinely be using this as my primary code editor, since it is so basic in functionality that it works really well (buttery smooth). Essentially I have everything that I need from what is essentially a fancy text editor and for anything else I'll just use the terminal. If you want to give it a spin, I've made a release at HyperRays/teditor only for macos arm64, but I have also tested it on windows and linux, just no binaries yet. Though you can spin up your favorite LLM and it'll probably be able to build it for you (It is essentially a rust project with some c dependencies and it needs zig due to ghostty).

🎉 We're starting an interop lab! p2ppsr/interop - Wallets + Applications - Transaction processing - Proof acquisition - Overlays and sync All in scope. Everything evaluated openly for conformance with shared interop objectives. Integration testing evaluation today, certification / accreditation of implementations in the future! BSV Browser, Yours and Metanet Explorer are added. There's an integration suite and it will grow over time. @johncalhooon @deggen @BGCryderman @shruggr we're looking forward to your contributions! We'll find the best ways to govern it over time, meritocracy led, impartial. It's Apache2 license.

@Dmardaytripper Hi awesome work! Can you do Descent 1 for 32x? I figured there's a chance...OpenSourcedGames/Descent