The Git Times

The asgeirtj/system_prompts_leaks repository continues to document system prompts from major AI models, with recent updates showing shifts in model behavior. On June 18, 2026, prompts were added for GitHub Copilot for macOS, Claude Design, and GPT-5.5 Codex. Earlier in June, Claude Fable 5’s prompt was published alongside a diff highlighting changes from Claude Opus 4.8. The repo also includes prompts for tools like Claude Code’s Glob and Grep functions, Perplexity Computer, and VS Code Copilot Agent. Maintained in JavaScript, the project serves as a reference for developers seeking to understand how model instructions influence output. By exposing these usually-hidden directives, it enables prompt engineers to fine-tune interactions and anticipate model constraints. The regular updates reflect the rapid iteration of AI systems, making the repo a practical tool for tracking real-time changes in model governance and design.

GitHub’s spec-kit project released version 0.11.4, introducing a Tasks to GitHub Project extension that bridges Spec-Driven Development with issue tracking. The update enables automatic creation of GitHub Project tasks when workflows execute, linking spec outcomes directly to engineering backlogs. This addresses a gap in translating executable specifications into actionable work items without manual handoff.

The release also includes hardened command registration, improved .gitignore patterns for lib/ and lib64/ directories, and preservation of argument hints in preset-generated SKILL.md files. A new specify bundle command simplifies role-based setup distribution. Linear integration was updated to v0.7.0, enhancing cross-platform task sync.

These changes refine the toolkit’s core promise: turning specifications into executable artifacts that trigger downstream processes. By embedding project management actions within the spec execution flow, teams reduce context switching and maintain traceability from product intent to code delivery.

The latest release of Open WebUI introduces oikb, a standalone knowledge base synchronization tool designed to streamline offline AI deployments. Built as a companion utility, oikb enables incremental syncing of local directories, GitHub repositories, S3 buckets, Confluence spaces, and over 40 other data sources into Open WebUI’s knowledge base. Only new or modified files are transferred, with automatic cleanup of removed files and orphaned subdirectories, preserving directory structure and showing per-file progress. This reduces bandwidth usage and avoids redundant reindexing in self-hosted environments. The tool supports checksum-based comparison and integrates directly with Open WebUI’s retrieval-augmented generation (RAG) pipeline, allowing users to maintain up-to-date contextual data without relying on cloud services. Targeted at enterprises and developers running LLMs locally via Ollama or OpenAI-compatible APIs, the update strengthens Open WebUI’s position as a secure, self-contained AI platform. Release v0.9.6 also includes smart directory sync for knowledge bases, enhancing usability for teams managing internal documentation, codebases, or research materials.

The Developer-Y/cs-video-courses repository has been updated with new video lecture playlists from the University of New South Wales (UNSW), including COMP6991 – Rust Programming (2022) and COMP1511 – Programming Fundamentals. These additions, pushed on June 21, 2026, mark the first significant content refresh since 2023 and reflect ongoing community curation of university-level computer science material.

The list, maintained since 2016, now indexes over 200 courses across 30+ specializations, from quantum computing to embedded systems. Contributors are directed to submit only full college/university-level course links via pull requests, excluding MOOCs, tutorials, or promotional content—a policy enforced to maintain academic rigor. Recent issue traffic shows active moderation against low-effort submissions, with spam accounts blocked per project guidelines.

This update reinforces the project’s role as a trusted, discipline-specific aggregator for educators, self-learners, and professionals seeking structured, video-based CS education outside commercial platforms.

Earthtojake has released version 0.3.5 of text-to-cad, a JavaScript library that equips AI agents with specialized skills for CAD modeling, robot description files, and hardware fabrication. The update refines the CAD viewer’s assembly inspection UX, fixes SRDF group state control loops, and improves orbit preview speed in the viewer. Core skills include generating STEP models from plain-language input, creating 2D DXF layouts, and exporting to STL, 3MF, and GLB formats. The library supports robotics workflows through URDF, SRDF, and SDF skill modules for defining robot structure, MoveIt planning groups, and simulator models. Additional skills handle part sourcing via STEP libraries, G-code slicing, and pre-upload validation for SendCutSend. Designed for local project integration, the toolkit enables agents to inspect, modify, and handoff CAD and robot artifacts without leaving the development environment. By focusing on interoperable formats and agent-ready workflows, text-to-cad bridges natural language instructions with mechanical engineering outputs, streamlining prototyping and automated design pipelines for hardware teams.

The ROS 2 documentation repository has enhanced its contributor workflow by integrating sphinx-autobuild for live-reload editing. Contributors can now run make serve to start a local server at http://localhost:2022 that automatically rebuilds and refreshes the browser upon saving changes to source files. This eliminates the need for manual make html cycles during documentation updates. The setup requires a Python virtual environment with pinned dependencies from constraints.txt, ensuring reproducible builds on Noble (Ubuntu 24.04). Spelling checks are available via make spellcheck. These tooling improvements lower the barrier for contributors to fix typos, clarify examples, or add new content to the official ROS 2 docs hosted at docs.ros.org.

The Maigret project has released a standalone Windows executable (maigret_standalone.exe) built from its main branch, enabling OSINT practitioners to run username-based investigations without installing Python or dependencies. The tool checks over 3,000 websites for account presence tied to a given username, extracting profile data, linked accounts, and other available information directly from web pages—no API keys required. Users can launch the executable via double-click or terminal, with options to save HTML reports (--html) or view full command-line help. The release includes recursive search capabilities, tag-based filtering (by site category or country), and mechanisms to detect and partially bypass blocks, CAPTCHAs, and censorship. Maigret fetches an auto-updated site database from GitHub once every 24 hours to maintain coverage. While the project is six years old and not new, this executable release lowers the barrier to entry for Windows users in security, red teaming, and digital investigations who need lightweight, offline-capable reconnaissance. The build is labeled as a development release and may be unstable; users are advised to consult the documentation for production use.

The latest release of RustScan, version 2.4.1, introduces a significant shift for developers: a standalone Rust library that allows direct integration of its port scanning capabilities into custom applications. Previously limited to command-line use, RustScan can now be invoked as a dependency via cargo add rustscan, enabling programmatic control over TCP and UDP scans, adaptive learning features, and result piping to Nmap or custom scripts.

This update responds to growing demand from security tool builders seeking lightweight, fast network reconnaissance without spawning external processes. The library maintains RustScan’s core promise—scanning all 65,535 ports in approximately three seconds—while eliminating overhead from process invocation. Benchmarks added in this release verify consistent performance across TCP and UDP protocols, with automated CI guarding against regressions.

Key improvements also include deduplication of IP addresses, fixed UDP timeout handling, and statically generated payloads for reliability. The project continues to avoid machine learning, relying instead on adaptive learning through usage-pattern math to refine scan order over time.

The projectdiscovery/httpx toolkit released version 1.9.0, introducing database output support for MongoDB, PostgreSQL, and MySQL to streamline result storage in security workflows. A new -fpt flag enables filtering responses by page type—such as login, captcha, or parked pages—using machine learning classifiers. Markdown output was also added for easier reporting. Bug fixes include graceful shutdown on interrupt to prevent data loss and improved resolver parsing. Written in Go and requiring Go 1.25+, httpx remains a fast, configurable HTTP probe tool used in bug bounty and penetration testing for host, URL, and CIDR scanning with automatic HTTPS-to-HTTP fallback. The project notes it is in active development with potential breaking changes.

Microsoft PowerToys version 0.100, released June 2026, delivers focused upgrades to its Windows utility collection. The update introduces a brand-new Shortcut Guide that visually displays active keyboard shortcuts in real time, aiding muscle memory and workflow efficiency. The Command Palette receives a significant overhaul with an integrated Extension Gallery, allowing users to discover and install community-built extensions directly, alongside new multi-monitor dock support for persistent access across displays. Power Display sees refinements in multi-monitor configuration handling and color profile management. Installation remains straightforward via .exe from GitHub releases, WinGet (winget install Microsoft.PowerToys), or the Microsoft Store, with per-user and machine-wide options. SHA256 hashes are provided for all installer variants to verify integrity. These iterative improvements reinforce PowerToys' role as a customizable, open-source toolkit for power users seeking granular control over their Windows 10 and 11 environments without modifying system core components.

The Go team released version 1.23 in June 2024, focusing on performance and WebAssembly advancements. Build times are reduced by up to 15% through improved caching in the go command and faster linkers. WASI preview2 is now enabled by default, allowing Go programs to run more efficiently in lightweight sandboxed environments like Cloudflare Workers and Fermyon Spin. The garbage collector received tuning for lower-latency applications, with reduced pause times in heap-heavy workloads.
The release also expands the go vet tool with new checks for unsafe pointer usage and misaligned struct access, aiding security audits. Module workflows gained stricter version validation to prevent dependency confusion.
These changes address growing demand for Go in edge computing and serverless platforms, where startup speed and resource efficiency are critical. The update maintains backward compatibility while strengthening Go’s position in infrastructure and cloud-native development.

The React Native team released version 0.86.0 on June 22, 2026, refining developer experience and framework reliability. Key updates include enhanced DevTools integration with a new custom track for React Native Renderer operations, enabling deeper performance tracing during app development. Animation backend improvements now use weak_ptr to prevent memory leaks when handling pullAnimationMutations in callback schedules, addressing a known stability issue in complex UIs. AccessibilityInfo gained missing Promise-based method tests and standardized null checks, improving screen reader compatibility. Dev Server updates allow Metro bundler to call global.__BUNDLE_LOADER_REPORTER__ during development builds, supporting custom tooling workflows. These changes target developers building production iOS and Android apps with shared JavaScript codebases, particularly those using Expo or custom DevTools plugins. The release maintains support for iOS 15.1+ and Android 7.0+, continuing React Native’s focus on bridging web-style React development with native performance.

The Embedded-Engineering-Roadmap project released version 1.2.3 on June 20, 2026, refining its curriculum for aspiring and practicing embedded engineers. This update focuses on bridging theory with hands-on firmware development, adding step-by-step guides for debugging ARM Cortex-M microcontrollers using JTAG and SWD interfaces. New sections cover real-time operating system (RTOS) task scheduling with FreeRTOS and Zephyr, alongside practical sensor fusion techniques for IMUs and environmental sensors using I²C and SPI protocols. The roadmap now includes annotated examples of bare-metal C code for STM32 and ESP32 platforms, emphasizing register-level configuration and interrupt handling. Contributors updated resource links to reflect current toolchains, including the latest versions of PlatformIO and VS Code extensions for embedded development. A new checklist helps learners validate project completeness — from schematic review to power consumption measurement — aligning with industry expectations for production-ready designs. The project maintains its structured progression from digital logic fundamentals to advanced topics like motor control and wireless communication, but now prioritizes measurable outcomes: each module ends with a deliverable, such as a working PID-controlled motor driver or a LoRaWAN-enabled environmental monitor. These changes respond to community feedback requesting less abstract theory and more verifiable, employer-relevant skills.

The ESPectre project released version 2.8.0, focusing on stabilizing its Wi-Fi spectre analysis (CSI) motion detection system. Key changes include switching the default traffic generator from DNS to ICMP ping, improving compatibility with routers that filter or rate-limit domain queries. Hampel outlier filtering is now enabled by default with a 5.0 MAD threshold, reducing false positives in noisy environments. The NBVI band selection process now evaluates four strategies per window and picks the lowest false-positive option, exposing scores for classic, entropy, and MAD-based methods. Detection logic was unified between firmware and Micro-ESPectre, using edge-driven binary motion publishing with configurable hit thresholds (default 3/3) and decoupled evaluation intervals. The machine learning detector, an experimental neural network feature, received updated training data grouped by chip type, a revised 9-feature model, and Hampel-aligned weights for better cross-device consistency. Movement Score output now uses a temperature-scaled gradient for smoother Home Assistant integration. All changes aim to improve plug-and-play reliability without requiring user calibration.

The BruceDevices/firmware project released version 1.15, introducing a NetCut-inspired ARP module for ESP32-based offensive security operations. This feature, contributed by @ryocuy, enables ARP spoofing and network disruption tools directly from the firmware, expanding its utility in red team engagements.

The update also integrates WDGoWars (wdgwars.pl) upload support for wardriving campaigns, allowing users to deploy and manage wireless reconnaissance scripts via the Bruce interface. Additional improvements include refined Karma attack handling, enhanced Flipper Zero-style app loading, and fixes for buffer overflow risks and UI scaling on small screens.

Bruce continues to support a wide range of hardware, including M5Stack Cardputer, LilyGo T-Deck, and Flipper Zero-compatible devices, with OTA and local flashing options via esptool.py or M5Burner. The firmware maintains dual-mode WiFi and BLE attack vectors, including Evil Portal, ARP poisoning, and Raw sniffing, alongside niche features like Brucegotchi and Pwnagotchi integration.

Official binaries are available through the project’s releases page, with web-based flashing via https://bruce.computer/flasher.

The Super Mario Bros Remastered project released version 1.1-26w21c, addressing a critical bug in its Godot 4.6-based level editor where room switching failed during map design. This fix restores core functionality for creators building custom levels using the project’s integrated editor, which supports importing original SMB1 ROMs and exporting shareable maps via Level Share Square. Built in GDScript, the remake faithfully recreates SMB1, The Lost Levels, Special, and All Night Nippon editions while enabling resource packs, custom characters, and portable mode via portable.txt. The update arrives nine months after launch, reflecting steady community engagement with 2,779 stars. Developers contribute through pull requests, adhering to the project’s open-source model that excludes original Nintendo assets. Users must supply their own legal SMB1 ROM to run the game, maintaining compliance with copyright. The editor now reliably handles multi-room level transitions, a foundational feature for complex map design. This patch resolves a regression that hindered iterative editing, directly impacting the usability of the custom level system highlighted in the README. No other changes were noted in the release notes, indicating a targeted fix for editor stability.

The Magpie project released version 0.12.1, refining its window upscaling tool for Windows 10 and 11 with a focus on usability and stability. Built in HLSL and leveraging DirectX 11, Magpie now includes an auto-hide cursor feature that conceals the pointer after a customizable idle delay, reducing visual clutter during scaled gameplay or application use. The update also overhauls the auto-scaling mechanism to prevent interruptions from pop-up windows, ensuring consistent scaling performance. Several compatibility fixes address long-standing issues: the scaled window now always stays on top (removing the user-toggle option), title bars are properly cropped during scaling, and monochrome cursor freezes have been resolved. A toolbar screenshot menu ID conflict was also patched. These changes build on Magpie’s core functionality—supporting fullscreen and windowed scaling with algorithms like Anime4K, FSR, and CRT shaders via a WinUI-based interface. The tool remains lightweight, multi-monitor aware, and requires Windows 10 v1903+ or Windows 11.

The Pyxel game engine, a long-standing tool for creating retro-style pixel art games in Python, has released version 2.9.6. This update refines its core loop for smoother frame timing and adds experimental WebGL export via Emscripten, letting developers build browser-compatible games without leaving Python. Sound handling now allows independent volume control per channel, addressing a frequent request from creators working on chiptune soundtracks. Built in Rust for performance but exposing a simple Python API, Pyxel maintains its 16-color palette and 4-channel audio limits to evoke 8-bit aesthetics. The engine includes integrated tools for image, sound, and tilemap editing, reducing reliance on external software. Despite being maintained by a single developer, the project shows steady activity, with the latest push in June 2024 indicating ongoing maintenance. Its MIT license and minimal dependencies continue to attract educators and hobbyists seeking constraints that focus creativity.