Design
Preset
Background
Text
Font
Size
Width
Account Pricing Friday, August 7, 2026

The Git Times

“Consider what the world would lose if each mind were to do its own indexing.” — Vannevar Bush

AI Models
Claude Opus 5 $25/M GPT-5.6 Luna $0.60/M Gemini 3.1 Pro Preview $12/M Grok 4.5 $6/M DeepSeek V4 Pro $0.87/M Qwen3.8 Max $6/M Kimi K3 $15/M
Full Markets →

mpv Switches to libplacebo GPU Backend for Better Wayland Support 🔗

Release 0.41.0 makes Vulkan hardware decoding preferred and adds native AAudio audio on Android

mpv-player/mpv · C · ▲ 41 in 1d Est. 2012 · Latest: v0.41.0

mpv, the long-standing command-line media player written in C, has shifted its default GPU rendering to the libplacebo-based gpu-next backend in version 0.41.

0, replacing the older gpu output. This change brings improved color management, better Wayland integration, and support for modern protocols like wp-color-representation-v1 and tablet input. The player now prefers Vulkan for hardware decoding when available, falling back only to non-copy variants before considering copy-based options.

Beyond graphics, the update adds clipboard writing support on both Wayland and X11, ambient light sensing via Linux sysfs ALS, and a new context_menu.lua script for right-click menus. On Android, mpv gains a JNI-free AAudio backend for lower-latency audio output. Windows users get helper scripts to register mpv as a system media app.

These changes reflect mpv’s ongoing focus on leveraging modern multimedia stacks while maintaining broad codec and format support through FFmpeg 6.1+. The player remains lightweight and configurable, appealing to developers and power users who prefer terminal-driven workflows or embedding mpv in custom tools.

The catch: Despite active development, mpv’s reliance on cutting-edge graphics APIs means older or low-power GPUs may experience tearing or stutter unless users manually enable performance profiles like --profile=fast, and hardware decoding is not enabled by default even when available.

Why this leads today mpv-player/mpv’s growing adoption reflects its critical role in developer workflows, CI pipelines, and automation where lightweight, scriptable media handling is essential — a foundational tool whose reliability drives practical impact across toolchains.

Use Cases
  • Developers testing video pipelines with scriptable CLI control
  • Linux users needing Wayland-native media playback with color accuracy
  • Android builders integrating low-latency audio in custom media apps

Source: mpv-player/mpv — based on the README and release notes.

Fresh on Hugging Face

Model Drops

The newest model releases builders are picking up right now.

More on the Front Page

LongHorizon-Harness Enables AI Agents to Complete Multi-Hour Desktop Tasks Reliably 🔗

The system separates planning, execution, and verification to preserve verified progress across context refreshes

AMAP-ML/LongHorizon-Harness · Python · 362 stars 3d old

LongHorizon-Harness lets AI agents like Claude Code and Codex operate across desktop applications and the CLI for extended periods without losing task state. It acts as an execution, state-management, and verification layer that runs on top of existing agents, ensuring only verified results enter persistent state.

By decoupling planning from execution and verification, the system prevents context overload and allows agents to resume complex workflows even after failures or context resets.

The harness supports both GUI and CLI workflows, enabling tasks to span environments — such as starting in a browser and continuing in a terminal — while maintaining durable, auditable progress. Version 0.1.2 introduces unified plugin management, stronger read-only auditor checks, role isolation, and improved process cleanup. It also expands diagnostic capabilities via the doctor command. One-command install gets teams running quickly, with native integration for Claude Code and Codex.

Recent activity shows strong momentum: the project reached #1 on Hugging Face Daily Papers weekly ranking shortly after release, and maintains rapid iteration with commits within the last day.

The catch: As an early-stage tool (v0.1.2) with only 5 open issues and 39 forks, real-world scalability beyond individual agent workflows remains unproven at enterprise scale.

Use Cases
  • Developers automating cross-app debugging sessions
  • QA teams running multi-step validation pipelines
  • Researchers executing long-form data analysis workflows

Source: AMAP-ML/LongHorizon-Harness — based on the README and release notes.

claude-red turns Claude into a context-aware red team operator 🔗

Each skill loads on demand via conversational triggers like mentioning SQLi

0xwilliamortiz/claude-red · JavaScript · 449 stars 1d old

claude-red provides structured SKILL.md files that prime Claude with expert offensive security methodology for specific attack surfaces, from SQLi to EDR evasion.

Skills activate contextually—mentioning a topic like wireless or ADCS loads the relevant skill without bloating the prompt. Designed for authorized red team work, bug bounty triage, and CTF prep, it lets operators explore techniques methodically. The project splits broad categories like Active Directory into 16 focused skills and mobile/IoT into 10, with roadmap-guided expansion. The catch: Most skills remain Windows-focused or lack proven cross-platform reliability in complex environments like hybrid cloud or RTOS firmware analysis.

Use Cases
  • Red teamers testing ADCS abuse during authorized engagements
  • Bug bounty hunters refining SQLi exploitation techniques
  • Security trainers teaching EDR evasion in controlled labs

Source: 0xwilliamortiz/claude-red — based on the project README.

MAGI-2 Preview Generates 10-Second Audio-Video Clips from Text or Image Prompts 🔗

Uses 6B active parameters per token via MoE architecture on 8 NVIDIA Hopper GPUs

SandAI-org/MAGI-2-preview · Python · 405 stars 2d old

MAGI-2 Preview is a 114B-parameter unified audio-video generation model that activates just 6B parameters per token. It generates 10-second clips with synchronized sound from text prompts (T2V) or text plus still image (I2V).

Inference runs in two stages: magi2_preview denoises at low resolution, then magi2_refiner upscales to 1080p. Requires Python 3.12, recent CUDA, ffmpeg on PATH, and 8 NVIDIA Hopper GPUs. Weights (~307 GB) must be downloaded separately into ckpt/ and are not bundled. The Docker image sandai/magi-2-preview includes dependencies; tag commits for reproducibility.
The catch: Generation speed is limited by denoising step count, as neither transformer has been step-distilled, making wall-clock time high despite parameter efficiency.

Use Cases
  • Researchers testing efficient video generation architectures
  • Developers prototyping text-to-video with synchronized audio
  • Teams evaluating MoE scaling for multimodal models on Hopper GPUs

Source: SandAI-org/MAGI-2-preview — based on the project README.

Spring Framework patches 16 CVEs in maintenance release v7.0.8 🔗

Critical WebSocket and WebFlux security flaws fixed amid rising AI-era threats

spring-projects/spring-framework · Java · 60.2k stars Est. 2010

The Spring Framework released v7.0.

8, addressing 16 CVEs including predictable session IDs in WebSocket modules and cross-site scripting via JSP form tags. Maintained by VMware’s Spring team, the project remains central to Java enterprise development with 38,803 forks and active GitHub Actions CI. Despite its maturity, the framework’s rapid patch cycle reflects ongoing vulnerability exposure in widely used components. Builders rely on it for dependency injection, MVC, and reactive programming across cloud-native systems. The catch: Its extensive abstraction layers can complicate debugging and slow startup times in resource-constrained environments.

Use Cases
  • Enterprises building scalable microservices with Spring Boot
  • Financial systems requiring transaction management and security
  • Java applications needing RESTful APIs and WebSocket support

Source: spring-projects/spring-framework — based on the README and release notes.

AI Agents Evolve Into Specialized, Composable Workforce Engines 🔗

Open source projects now build agent skills, memory, and audit tools for reliable, long-horizon task automation

Trendai-agents
KKKKhazix/human-writingAMAP-ML/LongHorizon-HarnessopenJiuwen-ai/jiuwenswarmmakecindy/cindyTencentCloud/TencentDB-Agent-Memory+30 more

The open source AI agent landscape is shifting from experimental prototypes to modular, production-ready components. Projects like AMAP-ML/LongHorizon-Harness enable agents to sustain complex workflows across desktop and CLI environments by preserving task state and enabling recoverable progress.

Memory systems such as TencentCloud/TencentDB-Agent-Memory centralize agent knowledge into reusable assets—chat history, skills, code graphs, and LLM-Wikis—governed and shared across frameworks. Skill marketplaces are emerging: huangruiteng/loopx provides a durable execution kernel for long-running agent teams, while ifixai-ai/iFixAi offers rapid independent auditing to verify agent behavior in under 120 seconds. Specialized skills address niche needs: ayghri/i-have-adhd prevents coding agents from obscuring answers, and blader/humanizer strips AI telltales from text. Agent orchestration is maturing via basketikun/infinite-canvas for multi-agent creative workflows and calesthio/OpenMontage, which turns coding agents into full video studios with 52 tools and 500+ skills. Even IDEs are becoming agent-native, as seen in ymichael/bb and rocketride-org/rocketride-server, which integrate agent orchestration directly into development environments.
The catch: Despite rapid innovation, the ecosystem remains fragmented—skills, memory formats, and agent protocols lack standardization, making cross-project integration brittle and often requiring custom adapters, which risks reinventing the very silos these tools aim to overcome.

Use Cases
  • Developers automate cross-app debugging sessions
  • Teams share verified agent skills across projects
  • Creators orchestrate multi-agent video production pipelines

AI Agent Skills Ecosystems Accelerate Specialized Workflow Automation 🔗

Open source projects are building modular, reusable skill libraries and agent harnesses for targeted technical tasks

Trendllm-tools
0xwilliamortiz/claude-redAMAP-ML/LongHorizon-Harnessmakecindy/cindyOcyss/boss-helperzhaoxuya520/reverse-skill+19 more

A clear pattern is emerging in open source around llm-tools: the rise of domain-specific skill ecosystems designed to extend AI agents into precise, repeatable workflows. Projects like 0xwilliamortiz/claude-red package offensive security techniques as `SKILL.

mdfiles that prime Claude with expert methodologies for SQLi, shellcode, or EDR evasion. Similarly,virgiliojr94/book-to-skillconverts technical PDFs into ready-to-use Claude Code skills, turning static knowledge into active agent capabilities. Beyond skills, agent orchestration is maturing.AMAP-ML/LongHorizon-Harnessenables AI agents to run complex, long-duration tasks across desktop apps and CLI with durable state, verified progress, and recovery—critical for unattended automation.huangruiteng/loopxprovides a lightweight state kernel for agent loops, supporting handoffs and quota-aware waking across Codex and Claude Code. Integration layers are unifying access.diegosouzapw/OmniRouteoffers a single endpoint to 290+ model providers with token-saving compression and auto-fallback, whileTencentCloud/TencentDB-Agent-Memorycreates team-level memory hubs (Chat Memory, Skill, LLM-Wiki, Code-Graph) for shared agent context. Meanwhile,bradautomates/claude-video` gives agents multimodal input by processing video frames for Claude analysis.
The catch: This ecosystem risks fragmentation—skills and harnesses often target specific agents (Claude Code, Codex) with limited cross-compatibility, and many remain early-stage prototypes lacking rigorous validation for production security or reliability-critical use cases.

Use Cases
  • Security researchers automate exploit development using Claude-red skills
  • Job seekers run end-to-end AI-driven application workflows locally
  • Teams share persistent agent memory across projects and frameworks

Web frameworks fuse AI, mobile, and zero-deps for rapid prototyping 🔗

New tools prioritize instant usability, offline capability, and unified toolchains over traditional stack complexity

Trendweb-frameworks
imsai-sh/zhuzhiliaoyuhuangerdi/InduSecAgentfastrepl/anarlogmakecindy/cindyzalando/skipper+17 more

A clear pattern emerges in open-source web frameworks: blending AI agents, mobile-first design, and zero-dependency architectures to slash setup friction. Projects like imsai-sh/zhuzhiliao deliver single-file, mobile-optimized web toys with real audio sampling—no build steps.

makecindy/cindy and EKKOLearnAI/hermes-studio offer out-of-the-box AI agents with web dashboards for chat, job scheduling, and analytics, eliminating config overhead. voidzero-dev/vite-plus unifies runtime, package management, and frontend tooling in Rust, aiming to replace fragmented JS chains. Meanwhile, nesquena/hermes-webui and zarazhangrui/frontend-slides enable mobile-accessible AI agent control and code-generated slides directly in-browser. Even non-framework tools like brightdata/cli and projectdiscovery/httpx reflect this ethos—terminal-first, single-binary utilities for web tasks. The shift favors immediacy: developers seek tools that work now, on any device, with minimal installation. AI integration isn’t tacked on; it’s core to the UX, as seen in AOrbitron/Eridanus’ LLM-driven function calling for desktop pets. This isn’t just about speed—it’s about reducing cognitive load so creators focus on ideas, not plumbing.

The catch: Many of these tools trade maturity for novelty—single-file demos scale poorly, AI agents hallucinate in workflows, and unified toolchains like Vite+ lock users into new ecosystems before proving long-term viability. Fragmentation risks loom as AI-web hybrids diverge in data models and plugin systems, threatening the portability they promise.

Use Cases
  • Developers build mobile-first web prototypes with zero dependencies
  • Teams deploy self-hosted AI agents with built-in session management
  • Engineers scrape and route web data via terminal-native tools

Quick Hits

jiuwenswarm JiuwenSwarm brings large language model intelligence to your daily apps via an AI agent built on openJiuwen, enabling seamless, context-aware interactions across platforms. 2.2k
MEV-Ethereum-Trading-Bot This Solidity-based MEV arbitrage bot automates profitable trades on Ethereum by connecting smart contracts to external scripts for real-time opportunity detection. 2.4k
InduSecAgent InduSecAgent is a Vue-powered security-focused AI agent designed to detect and respond to industrial system threats with real-time monitoring and adaptive defense mechanisms. 382
winget-pkgs The winget-pkgs repository hosts community-maintained Windows Package Manager manifests, enabling reliable, one-command installation and updates of software across Windows systems. 10.9k
dns-blocklists hagezi/dns-blocklists provides curated, regularly updated DNS blocklists to filter ads, trackers, and malware at the network level for a cleaner, safer browsing experience. 25.1k
anarlog anarlog offers an open-source TypeScript alternative to Granola AI, delivering intelligent meeting transcription and summarization with privacy-first, self-hostable architecture. 9k
gathered-scenes-zine-skill 670
Beyond GitHub

The AI Wire

What builders are reading today — the headlines, papers, and announcements that aren't trending repos.

From the labs & arXiv

OpenClaw’s npm plugin fix resolves update failures for TypeScript assistants 🔗

Patch enables singleton-array metadata handling in newer npm clients for plugin workflows

openclaw/openclaw · TypeScript · ▲ 113 in 1d 8mo old · Latest: v2026.7.1-2

OpenClaw’s latest release, v2026.7.

1-2, includes a targeted fix for npm plugin updates that previously failed when newer npm clients passed singleton-array metadata. The patch (#108336) ensures compatibility with official plugin installation and correction releases, addressing a silent blocker in the pnpm-based workflow.

The assistant runs locally via a Gateway that connects models, tools, and messaging channels like WhatsApp, Slack, and Signal through a unified control plane. It provisions Node.js runtimes automatically or uses existing installs (Node 22.22.3+, 24.15+, or 25.9+), with onboarding verifying model access and launching the Control UI for validation.

Tools execute on the host by default unless sandboxed, and the system treats inbound messages as untrusted input—a critical security posture for personal AI deployments. Companion apps extend functionality to voice, camera, and screen interactions across macOS, Linux, and Windows.

Despite active development—5,534 open issues and 81,018 forks—the project remains a single-operator focused tool, not designed for multi-user or enterprise-scale orchestration.

The catch: OpenClaws security model assumes local, single-user trust; exposing the Gateway remotely or enabling multi-user channels requires careful sandboxing and pairing approvals, which are opt-in and not enabled by default.

Previously in The Times “covered” — Aug 2

Use Cases
  • Developers testing local AI assistants on personal devices
  • Power users integrating LLMs with desktop automation tools
  • Privacy-focused builders avoiding cloud-hosted agent services

Source: openclaw/openclaw — based on the README and release notes.

More Stories

AI Prompt Repository Reveals Widespread Model Extraction Risks 🔗

34,837 forks show developer demand despite 159 open security issues

x1xhlol/system-prompts-and-models-of-ai-tools · Unknown · 142.6k stars Est. 2025

The x1xhlol/system-prompts-and-models-of-ai-tools repository aggregates system prompts from 25+ AI coding assistants including Cursor, Devin, and VSCode Agent, exposing how easily internal model behaviors can be reverse-engineered. With latest activity showing commits within the last day and sustained forking, builders use it to audit prompt hardening or benchmark AI tool consistency.

The catch: 159 unresolved issues indicate many prompts may be outdated or incorrectly attributed, risking flawed security assessments.

Previously in The Times “covered” — Aug 1

Use Cases
  • Security teams testing AI tool prompt injection defenses
  • Developers comparing reasoning patterns across coding assistants
  • Researchers studying emergent behaviors in LLM system prompts

Source: x1xhlol/system-prompts-and-models-of-ai-tools — based on the project README.

Prompt Library f/prompts.chat Adds MCP Server Support for AI Tool Integration 🔗

Enables seamless use of curated prompts as context in local and remote AI assistants

f/prompts.chat · HTML · ▲ 33 in 1d Est. 2022

The f/prompts.chat project now supports deployment as an MCP (Model Context Protocol) server, allowing developers to inject its curated prompt library directly into AI tools like Claude Desktop or custom agents.

This update lets users access over 143k stars’ worth of community-vetted prompts without manual copying, streamlining prompt engineering workflows. Self-hosting remains an option via Docker or manual setup with PostgreSQL, preserving privacy for enterprise use. The MCP integration works both locally and remotely, with configuration guided by updated documentation.
The catch: MCP support requires familiarity with AI tool extensibility, posing a barrier for non-technical teams seeking plug-and-play prompt access.

Previously in The Times “covered” — Aug 2

Use Cases
  • Developers inject prompts into Claude Desktop for consistent AI responses
  • Enterprises self-host prompt libraries with SSO and custom branding
  • Educators use the interactive book to teach prompt engineering concepts

Source: f/prompts.chat — based on the project README.

Quick Hits

spaCy spaCy delivers industrial-strength NLP in Python, enabling fast, accurate text processing for real-world applications like entity recognition and dependency parsing. 33.8k
machine-learning-for-trading This Jupyter notebook provides end-to-end machine learning for trading code — from data sourcing and feature engineering to model training and live execution — based on the 3rd edition textbook. 20.3k
generative-ai-for-beginners Learn to build generative AI applications with 21 hands-on lessons covering prompt engineering, LLMs, and integration — ideal for beginners entering the field. 116.9k
paperless-ngx Paperless-ngx is a self-hosted, community-driven document management system that scans, indexes, and archives your files with OCR and tagging for effortless retrieval. 44k
cookbook The Gemini API cookbook offers practical Jupyter notebook examples showing how to integrate Google’s Gemini models for text, vision, and multimodal tasks. 17.6k
julia Julia is a high-performance programming language designed for scientific computing, combining the speed of C with the usability of Python for numerical and data-intensive workloads. 49k

Openpilot Adds Driver Monitoring and Thermal Upgrades in v0.11.1 Release 🔗

New model improves driver attention tracking while thermal policy prevents comma four overheating during extended use

commaai/openpilot · Python · ▲ 13 in 1d Est. 2016 · Latest: v0.11.1

commaai/openpilot released v0.11.

1 with a refreshed driver monitoring model that uses updated neural networks to better detect drowsiness and distraction via the in-car camera. The release also overhauls the image processing pipeline for clearer visual input under low-light conditions, a common gap in prior versions. Thermal management for the comma four hardware received a targeted update to sustain performance during long drives in hot climates, addressing user reports of throttling. Support expanded to include Acura MDX 2022-2024 and Rivian R1S/R1T 2025 models, broadening compatibility beyond legacy platforms. The system continues to enforce ISO26262 safety standards through software-in-the-loop tests on every commit, with critical safety logic isolated in the panda module written in C. Developers can still run master branches directly, though prebuilt releases remain recommended for stability. Despite active development, the project maintains its core requirement: a comma four device and supported vehicle harness, limiting plug-and-play flexibility for alternative hardware.

The catch: Openpilot’s reliance on proprietary comma four hardware restricts deployment to users willing to invest in the specific sensor suite, excluding DIY builds on common platforms like Raspberry Pi or generic cameras.

Previously in The Times “covered” — Aug 5

Source: commaai/openpilot — based on the README and release notes.

More Stories

ROS2 Control Framework Maintains Steady Development Amid Growing Adoption 🔗

Last commit just hours ago shows active maintenance despite 8.9-year project age

ros-controls/ros2_control · C++ · 967 stars Est. 2017

The ros-controls/ros2_control project provides a generic C++ framework for robot control in ROS 2, supporting distributions from Humble to Rolling. Recent activity includes a commit within the last day, indicating ongoing maintenance despite the project's near-decade age.

Developers use it to build real-time controllers for robotic arms, mobile bases, and custom hardware interfaces across industrial and research robots. The framework abstracts hardware interactions, enabling reuse of controller code across different robot platforms. Documentation covers API references for all supported ROS 2 distros, with Docker images available for source and release builds. The catch: With 129 open issues, the project faces a backlog that may delay bug fixes and feature requests for newer ROS 2 versions.

Use Cases
  • Robotics engineers implement joint trajectory controllers
  • Researchers develop custom hardware interfaces for ROS 2
  • Industrial teams deploy reusable control logic across robot fleets

Source: ros-controls/ros2_control — based on the project README.

ROS 2 Documentation Repository Enables Live Preview of Docs Changes 🔗

Developers can now test documentation updates locally with automatic browser reload using sphinx-autobuild

ros2/ros2_documentation · Python · 971 stars Est. 2018

The ros2/ros2_documentation repo powers the live ROS 2 docs at docs.ros.

org, rebuilt nightly from source. Recent activity shows contributors using make html and sphinx-autobuild to preview changes instantly, with the serve target binding to 0.0.0.0:2022 for devcontainer access. The repository enforces reproducible builds via pinned Python versions in constraints.txt and supports multiversion testing for rolling releases. Contributors validate spelling with make spellcheck and manage custom corrections via codespell_whitelist and codespell_dictionary files. The catch: Over 190 open issues indicate ongoing maintenance challenges in keeping pace with ROS 2's rapid evolution.

Previously in The Times “covered” — Jul 31

Use Cases
  • Documentation writers previewing changes locally
  • Dev teams validating ROS 2 release notes
  • Contributors fixing spelling in API guides

Source: ros2/ros2_documentation — based on the project README.

Webots R2025a Adds ROS 2 Improvements and New Robot Models 🔗

Release includes pre-built binaries for Windows, Linux, and macOS with Docker support

cyberbotics/webots · C++ · ▲ 4 in 1d Est. 2018

Cyberbotics released Webots R2025a, updating its open-source robot simulator with new robot models, expanded demo scenes, and enhanced ROS 2 integration. The update provides pre-compiled packages including a Windows installer, Ubuntu .

deb and tar.bz2 files, a Snap package, and a Docker image via Hub. Users can simulate autonomous vehicles, robotic arms, and multi-robot systems using physics-based rendering and sensor plugins. The simulator supports C++, Python, and MATLAB interfaces for controller development. Despite active development, the project maintains 230 open issues, indicating ongoing challenges in stabilizing complex physics and sensor simulations across platforms.
The catch: High-fidelity simulations demand significant CPU and GPU resources, limiting real-time performance on consumer-grade hardware.

Previously in The Times “covered” — Aug 1

Use Cases
  • Researchers test autonomous navigation algorithms in urban environments
  • Educators teach ROS 2 concepts using simulated robotic arms and mobile bases
  • Engineers validate sensor fusion pipelines for drone swarm coordination

Source: cyberbotics/webots — based on the README and release notes.

Quick Hits

kornia Kornia provides a Python-based geometric computer vision library for Spatial AI, enabling robust image transformations and spatial reasoning in robotics and vision systems. 11.3k
verhoBot VerhoBot is a rope-free, pulley-less curtain-moving robot that automates opening and closing curtains to enhance sleep quality and integrate natural light into morning routines. 66
URDF-Studio URDF-Studio is a web-based visual robot modeler with 3D workspace, structured workflows, motor library integration, MuJoCo export, and AI-assisted URDF creation. 450
rosclaw Rosclaw delivers self-evolving runtime infrastructure for Physical AI, grounding agents in robot bodies via e-URDF, safety sandboxing, skill evolution, and physical memory for embodied intelligence. 178
nicegui NiceGUI lets developers build beautiful, interactive web UIs entirely in Python — no HTML, CSS, or JavaScript required — for rapid prototyping and deployment. 16.1k

HackTricks Wiki Adds Japanese Search Index to Boost Global Pentester Access 🔗

Latest release automates ja-language documentation retrieval for faster offensive security workflows

HackTricks-wiki/hacktricks · CSS · ▲ 7 in 1d Est. 2020 · Latest: searchindex-ja

The HackTricks-wiki/hacktricks repository released searchindex-ja, an automated build generating a Japanese-language search index for its CTF and pentesting knowledge base. This update enables users to quickly locate techniques in Japanese within the locally served wiki, accessible at http://localhost:3337 when running STM Cyber’s companion tool.

The project aggregates real-world exploit methods, CTF solutions, and research notes into a searchable offline resource, now enhanced with ja-language parsing via SerpApi-powered scraping and indexing. Contributors continue to maintain the core CSS-styled wiki, which references tools like PEAS and integrates with platforms such as Intigriti for bug bounty workflows and Modern Security for AI security training. The release notes confirm the index is rebuilt automatically, reducing manual maintenance for Japanese-speaking security practitioners. Despite six years of activity, the project relies on community-driven updates and lacks formal versioning for its knowledge base, meaning consistency across translations depends on contributor diligence.
The catch: The wiki’s effectiveness hinges on timely community contributions; outdated or missing techniques in non-English languages can create coverage gaps, requiring users to cross-reference with primary sources.

Previously in The Times “covered” — Aug 6

Use Cases
  • Pentesters searching Japanese CTF write-ups offline
  • Security trainers localizing offensive security materials
  • Bug bounty hunters verifying ja-language target recon methods

Source: HackTricks-wiki/hacktricks — based on the README and release notes.

More Stories

Strix AI Pentester Integrates with GitHub Actions for CI/CD Security 🔗

Autonomous agents now auto-scan pull requests and block vulnerable code before production deployment

usestrix/strix · Python · ▲ 303 in 1d Est. 2025

The open-source Strix project released v1.5.

0 with tighter GitHub Actions integration, enabling developers to run AI-powered penetration tests directly in CI/CD pipelines. Using Docker and an LLM API key, Strix deploys autonomous agents that dynamically test applications, validate findings with working exploits, and generate patches. It supports multi-agent orchestration for scalable testing and outputs compliance-ready reports. The tool aims to replace manual pentesting and reduce false positives from static scanners by focusing on real, exploitable vulnerabilities.
The catch: Strix requires a running Docker environment and a paid LLM API key, which may limit adoption in air-gapped or cost-sensitive environments.

Previously in The Times “covered” — Aug 3

Use Cases
  • Security teams automate vulnerability detection in web apps
  • Developers block insecure code via PR-based pentesting
  • Bug bounty hunters generate validated PoCs for faster reporting

Source: usestrix/strix — based on the README and release notes.

Shannon AI Pentester Adds Generic API Key Support for Any LLM Provider 🔗

v2.3.0 removes vendor lock-in by accepting SHANNON_AI_API_KEY for custom model integration

KeygraphHQ/shannon · TypeScript · ▲ 23 in 1d 10mo old

KeygraphHQ's Shannon open-source AI pentester now accepts a generic SHANNON_AI_API_KEY environment variable, enabling integration with any large language model provider beyond Anthropic, OpenAI, xAI, or AWS. Released August 6, v2.

3.0 shifts Shannon from hardcoded provider support to a flexible CLI flag, letting teams plug in local LLMs or niche APIs. The tool still analyzes source code, runs real exploits via browser automation, and reports only vulnerabilities with working proof-of-concept. Docker and Node.js 18+ remain required for the worker container and npx workflow. The catch: Shannon's effectiveness depends entirely on the quality and security awareness of the connected LLM, introducing variability in test depth and false positive rates across providers.

Previously in The Times “covered” — Aug 2

Use Cases
  • Security teams testing internal APIs with self-hosted LLMs
  • Developers validating fixes against OWASP Top 10 in CI pipelines
  • Red teams conducting gray-box assessments on staging environments

Source: KeygraphHQ/shannon — based on the README and release notes.

Quick Hits

berty Berty enables secure, resilient peer-to-peer messaging that functions offline or without trusted networks, ideal for privacy-focused builders needing censorship-resistant communication. 9.3k
algo Algo simplifies deploying a personal, auditable VPN in the cloud with one command, giving builders full control over encrypted traffic without vendor lock-in. 30.3k
httpx httpx is a high-speed, modular HTTP toolkit for Go developers that automates reconnaissance with retryable probes, making web asset scanning efficient and reliable. 10.2k
caldera Caldera automates adversary emulation using the MITRE ATT&CK framework, letting builders test defenses with realistic, repeatable attack scenarios. 7.2k
cve CVE aggregator continuously collects and updates the latest vulnerability data with PoC links, helping builders stay ahead of exploits in real time. 8k
Decepticon Decepticon is an autonomous Python-based hacking agent that performs end-to-end red team operations — from reconnaissance to exploitation — without human intervention. 5k

Lightpanda’s headless browser runs AI agents without Chromium bloat 🔗

Built in Zig, it exposes CDP for Puppeteer/Playwright while avoiding WebKit or Blink forks

lightpanda-io/browser · Zig · ▲ 30 in 1d Est. 2023 · Latest: nightly

Lightpanda is a headless browser engineered from scratch in Zig, not derived from Chromium or WebKit, to serve AI-driven automation. It implements the Chrome DevTools Protocol (CDP) to work seamlessly with existing tools like Puppeteer and Playwright, letting developers control it via standard scripts while avoiding the memory and startup overhead of full browser engines.

The project provides nightly builds for Linux (x86_64/aarch64) and macOS, plus Docker images for amd64 and arm64, with WSL2 support for Windows users. Benchmarks show it can request 933 real web pages on an AWS EC2 m5.large instance, highlighting its efficiency for large-scale scraping or agent workflows. Installation via Homebrew or AUR delivers the latest nightly, and the binary includes verification steps to ensure integrity. Lightpanda also features a native agent mode where users describe tasks in plain English or slash commands to drive browsing behavior directly.

The catch: As a non-Chromium browser, it may render pages differently or lack support for cutting-edge web features, posing risks for automation targeting modern, SPA-heavy sites.

Previously in The Times “covered” — Aug 4

Use Cases
  • AI agents scraping e-commerce sites at scale
  • Developers testing web apps with Puppeteer scripts
  • Automation workflows needing low-memory headless browsing

Source: lightpanda-io/browser — based on the README and release notes.

More Stories

Agent-managed Rust exhibit runs without human hands 🔗

Ultraworkers' claw-code harnesses crabs to maintain museum artifact

ultraworkers/claw-code · Rust · 195k stars 4mo old

The ultraworkers/claw-code repository operates as an agent-managed museum exhibit, where Rust-based harnesses automate planning, execution, and preservation using Gajae-Code and LazyCodex frameworks. Crabs sustain the system while agents handle labeling and upkeep, per the project's philosophy of zero human intervention.

Last pushed five months ago, it sees active maintenance with a commit just one day old and over 109,000 forks. Builders can inspect the artifact via USAGE.md for CLI workflows, parity checks, and container-first setup, though ACP/Zed integration remains pending.
The catch: Despite frequent commits, the project explicitly disclaims production use, positioning itself as a conceptual exhibit rather than a deployable tool.

Use Cases
  • Developers studying agent-driven Rust automation patterns
  • Teams exploring zero-human-intervention CI/CD harness designs
  • Researchers examining emergent behavior in crab-sustained codebases

Source: ultraworkers/claw-code — based on the project README.

Electron updates memory API for Linux and fixes cross-platform window bugs 🔗

v43.3.0 adds `available` field to `process.getSystemMemoryInfo()` on Linux and resolves DPI-related sizing issues

electron/electron · C++ · ▲ 18 in 1d Est. 2013

Electron v43.3.

0 exposes Linux memory availability via process.getSystemMemoryInfo()'s new available field, pulling data from /proc/meminfo. It fixes BrowserWindow size corruption on Windows secondary monitors with mismatched DPI and corrects dialog.showOpenDialog default path handling on Linux. The patch also resolves a Dirent.parentPath undefined error in fs.readdir with withFileTypes: true inside asar archives and a UAF in protocol.registerStreamProtocol. Despite steady updates, Electron’s large C++/Chromium footprint increases binary size and startup overhead compared to lighter alternatives.
The catch: High resource usage from bundling Chromium and Node.js may deter performance-sensitive desktop apps.

Use Cases
  • Developers building Slack-style chat apps with web tech
  • Teams porting internal tools to Windows, macOS, and Linux
  • Creators using Electron Fiddle to prototype desktop features quickly

Source: electron/electron — based on the README and release notes.

Quick Hits

react-native React Native lets builders create high-performance native mobile apps using familiar React components and JavaScript/TypeScript. 126.3k
vaultwarden Vaultwarden offers a lightweight, self-hosted Bitwarden-compatible password manager in Rust with strong security and low resource usage. 65.1k
ollama Ollama simplifies running and experimenting with diverse LLMs locally via a clean CLI and API, supporting models like Kimi-K2.6 and Qwen. 178k
rclone Rclone provides a powerful, unified command-line tool to sync, copy, and manage files across 40+ cloud storage providers with encryption and bandwidth control. 59k
llama.cpp llama.cpp enables efficient, CPU-optimized inference of large language models in C/C++ with minimal dependencies and support for quantized models. 123k
memos Open-source, self-hosted note-taking tool built for quick capture. Markdown-native, lightweight, and fully yours. 62.1k

TuyaOpen v1.9.0 cuts battery drain with ultra-low-power WiFi mode 🔗

New DTIM support and hardware drivers expand edge AI-IoT device longevity

tuya/TuyaOpen · C · ▲ 3 in 1d Est. 2023 · Latest: v1.9.0

TuyaOpen’s latest release adds WiFi ultra-low-power DTIM mode, significantly reducing standby consumption for battery-powered devices via the new ultra_lowpower_demo example. The framework now supports additional T5AI platform boards — ZECTRIX_T5AI_NOTE_4, SPARKLEIOT_T5_DEV, and LCKFB — integrating GC0308 camera and SC7A20 IMU drivers for vision and motion sensing.

Linux users gain RK3576 board support, enabling the your_chat_bot application to run directly on the hardware. These updates deepen TuyaOpen’s cross-platform C/C++ SDK, which already enables ASR, KWS, TTS, and STT speech pipelines while connecting to LLMs like ChatGPT, Gemini, and Deepseek via Tuya Cloud’s multimodal AI workflows. Developers can build secure, OTA-updatable smart devices compatible with Google Home and Alexa using Bluetooth, Wi-Fi, or Ethernet.
The catch: Despite active development, 55 open issues and reliance on independently updated third-party submodules pose integration risks for long-term product stability.

Previously in The Times “covered” — Aug 5

Use Cases
  • Build battery-operated voice-controlled smart sensors
  • Create camera-equipped edge AI devices for home monitoring
  • Deploy chatbot-enabled Linux gateways for industrial IoT

Source: tuya/TuyaOpen — based on the README and release notes.

More Stories

Automotive Skills Suite Adds Confirmation Reviewer Skills for Structured Deliverables 🔗

76 builder skills now pair with matching reviewer skills that validate outputs via KPI dashboards and findings tables

jherrodthomas/automotive-skills-suite · Unknown · ▲ 2 in 1d 3mo old

The jherrodthomas/automotive-skills-suite project has expanded its library to include 76 confirmation reviewer skills, each designed to validate the Excel outputs of its corresponding builder skill. These reviewers generate visual dashboards with KPI tiles, charts, and findings tables, ensuring structured deliverables meet automotive standards like ISO 26262, ISO/SAE 21434, and AIAG-VDA quality frameworks.

All skills are installable .skill files triggered by natural language in Cowork or Claude Desktop, with downstream skills consuming upstream outputs as stable file-format contracts. The chain enforces traceability across the engineering lifecycle, from concept to V&V execution. The catch: The reliance on Excel as a stable contract may limit adoption in teams using model-based tools or cloud-native data pipelines that avoid spreadsheet intermediaries.

Previously in The Times “covered” — Aug 3

Use Cases
  • Functional safety engineers automate safety case documentation
  • Cybersecurity teams generate TARA and IR plans from threat models
  • Quality engineers streamline APQP/PPAP workflows with FMEA integration

Source: jherrodthomas/automotive-skills-suite — based on the project README.

Rust Android Auto proxy adds energy forecast and album art injection 🔗

v0.22.0 enables EV routing tweaks and metadata manipulation for head unit customization

aa-proxy/aa-proxy-rs · Rust · 449 stars Est. 2024

The aa-proxy/aa-proxy-rs project bridges wireless Android phones to USB head units for Android Auto, now featuring protocol version override and Vehicle Energy Forecast publishing via WebSocket in v0.22.

  1. Album art injection supports EV energy forecast integration with configurable text placement and automatic H.264-derived artwork polling. Built in Rust with io_uring and tokio backends, it offers MITM mode for DPI changes, media stream inspection, and per-vehicle UI overrides. The companion app enables real-time bandwidth monitoring and automatic reconnection.
    The catch: Open issues number 19, and MITM mode requires careful configuration to avoid breaking head unit compatibility or violating Android Auto’s terms of service.
Use Cases
  • DIY builders enable wireless Android Auto on legacy head units
  • EV owners inject forecast data into media metadata for routing awareness
  • Developers inspect decrypted AA streams via TCP for custom media pipelines

Source: aa-proxy/aa-proxy-rs — based on the README and release notes.

Eurorack Module Project Reorganizes Into Dedicated Repositories 🔗

Recent split improves accessibility for drum, filter, and VCO builders

Fihdi/Eurorack · Unknown · 139 stars Est. 2024

Fihdi/Eurorack has reorganized its schematics and PCBs into separate repositories for individual modules. As of May 2026, the 30hp Drum, Hi Hat, WAV Player (MiniDrumkit), 8hp ADSR (SCULPT), dual VCA/ringmod (BIPO), SVF12 filter, VCAR slew limiter, ATS-20 radio receiver, UNO VCO/LFO, and DICE S&H/random module now live in their own repos.

This split streamlines access for builders targeting specific functions like percussion or modulation. The project remains active, with the last push under a day ago and zero open issues.
The catch: Documentation for legacy modules may be fragmented during the ongoing reorganization, complicating cross-module system builds.

Use Cases
  • Drummers building custom 30hp percussion systems
  • Synth designers integrating SVF12 filters into modular setups
  • Experimenters sourcing UNO VCO/LFO schematics for pitch modulation

Source: Fihdi/Eurorack — based on the project README.

Quick Hits

espectre ESPectre detects motion using Wi-Fi signal analysis (CSI) and integrates with Home Assistant for smart home automation. 8.9k
litex LiteX lets builders design and synthesize custom hardware easily with Python-based SoC construction. 4k
photobooth-app Photobooth-app offers a free, open-source Python photobooth with a modern Vue3 frontend for instant fun snapshots. 305
firmware OpenIPC firmware provides a community-driven, open-source alternative for IP cameras with enhanced flexibility and control. 2.1k
ghdl GHDL simulates VHDL 2008/93/87 designs accurately, enabling reliable FPGA and ASIC verification in pure Python. 2.9k

HL7's SQL-on-Fhir bridges FHIR data to SQL engines for healthcare analytics 🔗

Recent commit shows active maintenance with new FHIRPath view definitions added

HL7/sql-on-fhir · GLSL · ▲ 1 in 1d Est. 2023

The HL7/sql-on-fhir project maintains the SQL on FHIR Implementation Guide, enabling developers to create portable, tabular views of FHIR data using FHIRPath expressions. These views map nested healthcare resources to SQL-compatible tables, allowing analysts to query patient records, observations, and claims with standard SQL tools instead of custom FHIRPath parsers.

The specification supports unnesting arrays, renaming columns, and filtering logic directly in view definitions, reducing boilerplate when integrating FHIR with data warehouses or BI platforms.
Recent activity includes a commit seven days ago updating view definitions in the input/pagecontent directory, suggesting ongoing refinement of the logical mapping between FHIR structures and SQL schemas. The project uses Sushi and HL7 IG Publisher to generate documentation, with npm scripts for local builds and continuous updates. Implementations like sql-on-fhir.js provide runtime engines that execute these views against FHIR servers.
The catch: The specification remains focused on read-only projections; write operations or transactional updates to FHIR data through SQL views are not defined, limiting use in applications requiring bidirectional data synchronization.

Previously in The Times “covered” — Aug 6

Use Cases
  • Healthcare analysts querying patient vitals with standard SQL
  • Data engineers building ETL pipelines from FHIR to data lakes
  • Researchers aggregating claims data across FHIR servers using BI tools

Source: HL7/sql-on-fhir — based on the project README.

More Stories

EnTT v4.0.0 drops legacy utilities, adds concepts for safer C++20 ECS workflows 🔗

Header-only library used in Minecraft and ArcGIS now requires C++20 for full feature set

skypjack/entt · C++ · ▲ 2 in 1d Est. 2017

EnTT’s v4.0.

0 release removes deprecated core utilities like entt::identity and entt::popcount, replacing them with C++20 concepts such as enum_bitmask and cvref_unqualified. The update refines type_info and component_traits to favor concepts over SFINAE, improving compile-time safety and reducing template bloat. Views now support view-to-view conversion across all types, and storage destruction sees performance gains for trivially destructible components. Despite its header-only, dependency-free design, the library’s shift to C++20-only features may exclude teams on older toolchains.
The catch: Adopting v4.0.0 requires a C++20-compatible compiler, limiting use in legacy game engines or constrained build environments.

Use Cases
  • Game studios implement entity-component systems for AAA titles
  • GIS developers build spatial data platforms with Esri’s ArcGIS Runtime
  • Indie creators prototype physics-based ragdoll systems in C++20 projects

Source: skypjack/entt — based on the README and release notes.

Rust GUI Library Egui Adds Mobile Keyboard Sync and Panel Drag Reopen 🔗

Latest release improves touch input and window theming consistency across platforms

emilk/egui · Rust · ▲ 12 in 1d Est. 2019

Egui’s 0.36.

0 update brings functional mobile keyboard support, fixing autocomplete, autocorrect, and IME handling on iOS and Android via eframe web. Users can now drag panel handles to reopen closed interfaces, a small but notable UX refinement. Window decorations now sync with app themes, eliminating OS chrome mismatches during dark/light mode switches. These changes refine egui’s immediate mode GUI for cross-platform apps, especially where touch and theme consistency matter. The library remains dependency-light but requires manual setup for native Linux backends.
The catch: Mobile improvements depend on eframe web, leaving native mobile integration less mature.

Previously in The Times “covered” — Aug 1

Use Cases
  • Game devs building cross-platform tools with Rust
  • Developers creating Wasm-based admin panels
  • Teams needing theme-consistent GUIs in desktop apps

Source: emilk/egui — based on the README and release notes.

Material Maker Adds UI Tweaks in Godot-Based Texture Tool Release 🔗

Version 1.7 improves panel layout, shortcuts, and file dialogs for procedural texture artists

RodZill4/material-maker · GDScript · 5.8k stars Est. 2018

RodZill4/material-maker, a Godot Engine-based tool for procedural texture creation and 3D model painting, released version 1.7 with UI refinements contributed by williamchange.

Updates include customizable panel layouts, side-panel toggling, quickbar shortcuts, and enhanced node graph interactions like seed rerolling and connection hints. File dialogs now show thumbnails and remember display modes, while single-window mode and improved editor titles streamline workflows. Built on GDScript, it uses Godot’s GraphEdit node system to node-edit textures and brushes.
The catch: Active development relies heavily on individual contributors, with 312 open issues indicating potential delays in bug fixes and feature requests.

Use Cases
  • Game artists create seamless textures for Godot projects
  • Designers paint 3D models using procedural brush workflows
  • Developers synthesize materials via node-based GLSL graphs

Source: RodZill4/material-maker — based on the README and release notes.

Quick Hits

Godot-Game-Template Maaack/Godot-Game-Template: A ready-to-use GDScript template with polished menus, scene management, and example gameplay to jumpstart your Godot 2D/3D projects fast. 1.6k
renodx clshortfuse/renodx: A powerful HLSL-based engine that enhances DirectX games with advanced post-processing, customizable visuals, and low-overhead rendering for modders and devs. 1.6k
bevy bevyengine/bevy: A refreshingly simple, data-driven Rust game engine that prioritizes ergonomics and performance for building modular, scalable games without boilerplate. 47.5k
engine playcanvas/engine: A blazing-fast WebGL/WebGPU-powered runtime enabling stunning 3D graphics, physics, and XR experiences directly in the browser with zero plugins. 16.4k
VulkanCppExamples myemural/VulkanCppExamples: A comprehensive C++20 tutorial collection demonstrating modern Vulkan rendering techniques — from basics to advanced — with clear, executable code samples. 123
The Git Times AI Desk
Ask about today's stories — or hit “Ask about this” on any article to focus on one.

Unlock the Git Times AI desk to ask about today's stories and the AI model market.

Upgrade to Premium
Answers by the Git Times AI desk · verify before you ship