The Git Times

Spec Kit released version 0.4.3 this week, focusing on consistency and platform reliability for its Spec-Driven Development toolkit.

The update migrates legacy dotted Kimi directories and unifies skill naming with Codex, as implemented in pull request #1971. This reduces maintenance overhead for community extensions and presets that rely on predictable directory structures when working with different AI agents.

A separate change fixes PowerShell 5.1 compatibility by replacing the null-conditional operator, detailed in pull request #1975. The adjustment ensures the specify CLI runs reliably on older Windows environments without syntax errors.

The toolkit continues to treat specifications as executable assets. Instead of serving as temporary documentation, they directly drive code generation through supported AI models. The specify init command scaffolds projects with flags such as --ai claude, while specify check validates the local environment.

These refinements arrive as more engineering teams integrate multiple AI agents into their workflows. Standardized naming reduces friction when switching between models, and broader shell support expands accessibility for enterprise developers constrained by corporate Windows images.

The release maintains the project's emphasis on predictable outcomes derived from clearly defined product scenarios rather than ad-hoc implementation.

Word count: 178

OpenClaw v2026.3.28 introduces async requireApproval to before_tool_call hooks, allowing plugins to pause execution and request explicit user consent. Approvals can arrive through Telegram buttons, Discord interactions, an exec approval overlay, or the universal /approve command on any connected channel.

The update migrates xAI tooling to the Responses API and adds first-class x_search support. The bundled Grok plugin now auto-enables based on web-search configuration, with optional x_search setup offered during openclaw onboard. MiniMax gains an image-01 provider for both image generation and image-to-image editing, including aspect-ratio controls.

Two breaking changes affect long-term users. The deprecated qwen-portal-auth integration has been removed; administrators must migrate to Model Studio API keys. Automatic config migrations for installs older than two months have also been dropped, so legacy keys now fail validation outright.

These changes reflect a maturing focus on controlled execution and provider hygiene while preserving the project's core offer: a single-user AI assistant that operates across WhatsApp, Slack, Discord, Signal, Matrix, Teams and 15 other platforms. It runs on Node 24 (or 22.16+), supports voice on macOS/iOS/Android, and renders live canvases under user control.

The release continues OpenClaw's emphasis on keeping the assistant local, fast, and always-on without relying on third-party orchestration layers.

Google has shipped version 0.35.3 of gemini-cli, applying targeted fixes that resolve cherry-pick conflicts from the prior release. The update forms part of the project's established weekly cadence, which delivers preview builds every Tuesday and stable versions shortly after.

The tool provides direct terminal access to Gemini 3 models that support a 1M token context window and enhanced reasoning. This capacity allows developers to work with entire repositories or lengthy technical specifications in one session.

Built-in capabilities include:

• Google Search grounding for factual accuracy
• Native file operations
• Shell command execution
• Web content retrieval

MCP (Model Context Protocol) support enables custom mcp-client and mcp-server implementations, letting teams connect the agent to internal services or proprietary data sources. The free tier supplies 60 requests per minute and 1,000 daily requests for personal Google accounts.

Installation remains flexible across environments, from npx @google/gemini-cli for instant runs to global npm, Homebrew, MacPorts, or conda-based setups. Under Apache 2.0 licensing, the project continues to evolve as a lightweight yet extensible bridge between command-line workflows and frontier models.

(178 words)

NiceGUI has introduced version 3.9.0, featuring new interface elements and improved integration capabilities for Python developers.

The release adds ui.parallax, based on Quasar, allowing developers to create dynamic scrolling backgrounds and layered content. This enhances the visual appeal of web-based applications without requiring frontend frameworks.

For three-dimensional visualizations, ui.scene now supports "trackball" and "map" camera controls. These additions facilitate more natural interaction with complex 3D models, beneficial for simulation and robotics applications.

Native mode gains support for window events including shown, resized and file drop through app.native. Developers can now respond directly to these system-level actions in their Python code.

Security receives attention with measures to prevent memory exhaustion from media streaming routes. This patch mitigates risks in applications handling user-uploaded or streamed content.

Multiple bug fixes address session storage conflicts with FastAPI, log scrolling issues in Firefox, table header animations and page navigation problems. Compatibility with PyInstaller has also been improved.

These updates come as Python continues to gain traction for full-stack development. NiceGUI's approach of running a webserver that the browser accesses, or operating in native desktop windows, provides flexibility for different deployment scenarios. The implicit reload on code changes speeds up the iteration process significantly during development.

openpilot v0.11.0 introduces a driving model fully trained inside a learned simulator, a shift that has produced measurable gains in longitudinal performance within Experimental mode.

The new model replaces earlier versions trained primarily on real-world data. According to the release notes, the simulator-trained approach has tightened acceleration and braking response, particularly in stop-and-go traffic. Standby power consumption on the comma four has also been cut by 77 percent to 52 mW, extending the time the device can remain connected without draining the vehicle battery.

Community contributors added support for the Kia K7 2017 and Lexus LS 2018, increasing the total number of officially supported vehicles beyond 300. Installation on a comma four remains straightforward: users point the device to openpilot.comma.ai during setup and attach the appropriate car harness.

The project continues to ship four main branches. The release-mici branch delivers stable updates, while nightly-dev includes experimental features for select models. All code is written in Python and developed openly on GitHub, with comma maintaining the core stack and individual contributors adding car-specific integrations.

The release reflects openpilot’s maturing architecture: an operating system for robotics that incrementally replaces factory driver-assistance software rather than bolting new hardware on top of it.

(178 words)

Proxmox VE administrators gained access to an additional deployment script with the project's most recent release. The update on March 30 added Bambuddy to the collection of supported applications and implemented a rename from BirdNET to BirdNET-Go.

This breaking change requires existing users to update any references to the old script name in their workflows. The adjustment reflects upstream developments in the bird identification software.

The community-scripts/ProxmoxVE repository offers Shell scripts for creating optimized LXC containers and virtual machines on Proxmox VE 8.4.x through 9.1.x. Supported operating systems include Debian, Ubuntu and Alpine, with ready-made scripts for Home Assistant, Docker, network tools, security solutions and smart-home services.

Installation has been streamlined through the web interface at community-scripts.org or by adding a local menu directly in the Proxmox UI. The local option uses this one-line command:

bash -c "$(curl -fsSL https://raw.githubusercontent.com/community-scripts/ProxmoxVE/main/ct/pve-scripts-local.sh)"

Community members continue to contribute new scripts and maintain existing ones, providing both simple and advanced configuration modes, automatic updates, and post-install troubleshooting tools. Regular security patches and performance optimizations keep the scripts relevant for production self-hosting and homelab environments. Active Discord and GitHub channels support users with feature requests, bug reports and implementation guidance.

Berty has released v2.471.2, addressing GitHub workflow issues and sustaining development of its mature zero-trust messaging platform. First created in 2018, the project continues refining a system designed to operate with or without internet, cellular coverage or any trusted intermediary.

The app delivers end-to-end encrypted messaging by default while collecting minimal metadata. Users create accounts without phone numbers or email addresses. Local discovery relies on Bluetooth Low Energy and mDNS, enabling fully offline conversations. When connectivity exists, libp2p, IPFS, and CRDTs provide distributed storage and conflict-free replication across peers.

The monorepo contains a React Native mobile client written in TypeScript targeting both Android and iOS, alongside Go components using gomobile. Developers can run berty mini for a lightweight CLI messenger or berty daemon to operate a full node exposing the Wesh Protocol API.

As governments increase network monitoring and impose shutdowns, Berty's serverless architecture offers concrete resilience. Messages remain encrypted and synchronized even on adversarial networks. The French nonprofit steward continues balancing usability with uncompromising decentralization, keeping the codebase fully open for audit and contribution.

The latest maintenance release, while technical, confirms the project's ongoing commitment to privacy tools that function precisely when conventional messengers cannot.

Astral has released uv 0.11.2, sharpening the self-update process in its Rust-based Python package and project manager. The uv self update command now fetches the manifest from the mirror first and uses uv’s own reqwest client, delivering clearer success and failure messages even when --quiet is specified.

On Windows, the update adds a dedicated PE editing error type, giving developers more precise diagnostics during binary operations. A preview feature evaluates extras and groups when determining auditable packages, strengthening supply-chain security checks. A separate fix removes redundant project configuration parsing for uv run, reducing overhead on repeated script execution.

These changes refine an already comprehensive tool that unifies dependency resolution, virtual environment management, and Python version handling. uv maintains a universal lockfile format, supports Cargo-style workspaces, and provides a pip-compatible interface that runs 10-100x faster than the original. Its global cache deduplicates packages across projects, conserving disk space on developer machines and CI runners alike.

The release continues uv’s steady evolution since its 2023 debut, focusing on operational robustness rather than flashy new features. Installation remains unchanged—via standalone curl or pip—and the binary can still update itself without external toolchains.

Key enhancements in 0.11.2

• Mirror-first manifest fetching for uv self update
• Dedicated Windows PE editing error reporting
• Preview support for extras and groups in audit paths

Meilisearch has introduced Dynamic Search Rules in version 1.41.0. The experimental feature lets developers promote specific documents based on query conditions or time windows without changing core indexes.

Rules trigger when queries contain defined substrings or during scheduled periods. Multiple documents can be pinned in exact order at any position in the results list. The system integrates cleanly with existing capabilities: filtering, pagination, facet distribution, hybrid search, and federated search.

Developers create and manage rules through the API. A rule is defined with PATCH /dynamic-search-rules/{uid} using JSON that specifies conditions and pinning actions. Rules are removed with DELETE /dynamic-search-rules/{uid}. The approach gives teams precise control over promotional content, seasonal highlights, or editorial priorities.

The Rust engine maintains its signature performance, typically returning results in under 50 milliseconds. It continues to combine semantic and full-text search while offering typo tolerance, geosearch, synonym support, and multi-tenancy controls.

This update addresses a practical need for context-aware result shaping that previously required application-layer workarounds. For teams already running Meilisearch in production, the new rules provide a native, declarative way to influence relevance without sacrificing speed or consistency.

The Linux kernel project has expanded its official documentation to explicitly address AI Coding Assistants using LLMs and related tools. This change reflects the growing presence of automated assistance in what remains the most scrutinized codebase in open source.

The updated README now lists AI roles alongside new developers, security experts, and hardware vendors. It directs them to the same foundational materials: the development process, patch submission rules, coding style guidelines, and core API documentation. All are expected to follow the established code of conduct and licensing terms in the COPYING file.

Recent commits, with the latest push on March 30 2026, show the tree remains under active maintenance. The addition of AI guidance acknowledges that large language models are already being used for initial code generation, bug triage, and documentation work, while preserving the human-led review process that has defined kernel quality for decades.

Security professionals and academic researchers retain dedicated sections on hardening, vulnerability analysis, and architectural internals. The kernel itself continues to manage hardware resources and provide the fundamental services that power everything from embedded devices to the world's largest AI training clusters.

This pragmatic update keeps the project aligned with contemporary development realities without relaxing its technical standards.

photobooth-app has updated to v8.7.0, its final feature release in the v8 series. The main addition is a new Rclone-based synchronization tool that replaces several older transfer methods.

The backend now depends on the rclone-bin-api package, which bundles current Rclone binaries for Windows, macOS, Linux x64 and ARM. This removes the need for manual Rclone installation and simplifies deployment on Raspberry Pi and desktop systems alike. Existing QR share, FTP and Nextcloud options remain for now but are expected to be phased out.

Other changes improve reliability. Services can now signal permanent crashes for illegal configurations, preventing repeated failed recovery attempts. Multicamera setups gain TCP keepalive support through pynng, fixing dropped connections during long stills sessions. The frontend renames the download portal to sharepage and includes minor visual tweaks alongside updated dependencies.

The Python application with Vue3 interface captures stills, animated GIFs, collages, boomerangs and 3D wigglegrams. It supports DSLR cameras via gPhoto2, Raspberry Pi Camera modules, webcams, and combinations of these. Live preview runs during countdowns, while WLED integration drives LED rings for visual feedback. 3D-printable enclosure designs and MIT licensing continue to support the DIY community.

**

HackRF has received a focused maintenance release that improves core radio performance. Version v2026.01.3 corrects mixer frequency lock failures that previously caused intermittent instability during transmission and reception. The fix delivers more predictable behaviour across the device's operating range, particularly important for applications requiring sustained carrier stability.

The update also grants access to a larger SPI flash on the HackRF Pro. This change allows users to store more complex firmware images and capture bigger datasets directly on the hardware without external memory cards.

The platform, written in C, continues to provide both open hardware designs and supporting software for software-defined radio work. Documentation remains available on Read the Docs, with source files in the repository's docs folder. Local PDF builds on Ubuntu require latexmk and texlive-latex-extra, followed by the standard make latex and make latexpdf sequence.

Contributors are encouraged to submit documentation improvements through pull requests. Users seeking help should first consult the troubleshooting page, then open GitHub issues. The project maintains Discord channels for discussion, though response times for technical support labelled issues average two weeks.

These incremental changes demonstrate the project's ongoing commitment to reliability for its established user base of RF engineers and researchers.

Kubernetes node management takes another step toward hardware agnosticism with the latest Node Feature Discovery release. Version 0.18.3 introduces official support for ppc64le and s390x, complete with dedicated container images from the Kubernetes registry.

This change allows operators of IBM Power and z/Architecture systems to automatically discover and label node features. The add-on scans for CPU capabilities, such as instruction set extensions, and applies structured labels for use in scheduling decisions.

Installation uses familiar tools. The Helm chart or Kustomize overlay now pulls the multi-arch images by default when using v0.18.3.

Example labels include feature.node.kubernetes.io/cpu-cpuid.ADX and similar entries for AESNI and other features. These enable fine-grained control over where pods run.

Additionally, the release corrects the "test" subcommand in the accompanying kubectl-nfd plugin, streamlining validation of feature rules.

The update matters as organizations increasingly deploy Kubernetes across varied server architectures. Whether in on-premises data centers or hybrid cloud setups, consistent feature detection simplifies operations.

Users can define custom feature rules using configuration files, extending beyond built-in detectors for CPUID and RDT. This flexibility has made the tool popular in high-performance computing and machine learning clusters where specific hardware instructions can significantly impact performance.

Dear ImGui has released version 1.92.6, the first major update following last year's 10th anniversary of v1.00. The new version refines existing systems rather than introducing sweeping changes, with particular attention to stability and feature discoverability for long-term users.

The library continues to output optimized vertex buffers that integrate directly into any 3D rendering pipeline. Its immediate-mode design eliminates the need to maintain separate UI state, reducing synchronization bugs that commonly appear in tool development. Developers report integrating basic functionality in roughly 25 lines of code within established codebases.

Core strengths remain unchanged: minimal dependencies, renderer agnosticism, and explicit focus on programmer tools rather than end-user applications. The release notes urge users to read the full changelog, noting that many teams overlook useful widgets added over the past decade.

Funding remains a central theme. The maintainers highlight invoiced sponsorship options for companies using the library in commercial engines and embedded systems, seeking sustainable support beyond individual donations. Multi-platform consistency across Windows, macOS, Linux and console environments keeps it relevant for teams shipping tools alongside their products.

The update reinforces Dear ImGui's position as reliable infrastructure rather than a flashy newcomer, delivering predictable performance where it is needed most.

Godot developers have a new reason to revisit their dialogue systems with the release of godot_dialogue_manager version 3.10.2.

The update introduces an option to generate static IDs for the current file. This helps maintain consistent line references as projects evolve, reducing errors in save systems and translations.

Another notable change allows for different indentation styles. Developers can now align the tool with their preferred code formatting without compromising functionality.

The release focuses heavily on stability with several fixes. These include improvements to the find function in the dialogue panel, the C# mutation handler, null checks for autoloads in symbol lookup, and character name completions that previously included commented lines.

A bug in the _check_condition() function was also resolved, preventing crashes with non-primitive values.

These enhancements matter because reliable dialogue tools are essential for the growing number of narrative-focused titles built with the Godot engine. The addon provides a stateless branching dialogue editor and runtime, allowing creators to write in a script-like manner and integrate easily into their games using GDScript or C#.

It supports advanced features such as conditions, mutations, and dialogue balloons. As Godot gains traction among indie developers, maintaining such specialized addons becomes increasingly important for efficient content creation.

The project continues to see contributions from the community, with a new contributor submitting a pull request in this release.

Flame, the established Flutter-based game engine, has released version 1.36.0, delivering a series of targeted fixes to improve stability and functionality.

Among the changes, the update corrects ray direction normalization drift that affected accuracy in ray casting operations. CircleComponent now properly initializes its center offset in the constructor, while IconComponent resolves fontPackage handling during rasterization.

Hitbox calculations have been updated to correctly factor in parent scale and rotation, addressing a significant source of bugs in games with dynamic objects. The buildContext becomes available during onLoad and onMount methods, allowing developers to access context earlier in the component lifecycle.

Text rendering sees improvements with added CJK wrapping support in TextBoxComponent. The release also prevents removed children from being erroneously re-added when their parent component is moved.

These enhancements build on Flame's core offerings, which include a complete game loop, a component/object system, collision detection, gesture and input handling, and support for images, animations, sprites and particles. The engine integrates seamlessly with Flutter, providing utilities that simplify game development.

Bridge packages extend its capabilities to include audio playback, state management with Bloc, and other integrations, making it a comprehensive solution for Flutter game developers.

Community resources include extensive documentation, examples, tutorials and an active Discord server for support.