Vollständiges Transkript anzeigen (1.576 Wörter)
Every week, developers release powerful open-source dev tools, and this weekly dev tool project update video brings them together in one place. Top trending open-source and best dev tool projects this week cover code knowledge graphs, reverse engineering tools, web crawlers, coding agent controllers, and memory layers. You'll discover useful and trending developer tools you can start using right away, explained fast and to the point. Without wasting time, let's get started.
Before we jump into today's project updates, here's a quick announcement for everyone. We've launched a brand new YouTube channel called AI Agents Studio, dedicated entirely to AI agent projects, tutorials, and tools. So if you're interested in staying up-to-date with the latest AI agent open-source projects, learning how to build your own agents, or exploring cutting-edge agent frameworks, make sure to check it out. Subscribe now to get weekly videos, in-depth guides, and real-time project breakdowns. The link is right there in the description, don't miss it. Alright, let's get into today's video.
Project number one, GitNexus, code knowledge graph for AI agents. Coding agents like Cursor and Claude Code often edit functions blind to everything that depends on them. GitNexus indexes a repo into a knowledge graph, parsing code with tree-sitter and resolving call chains, then exposes it over MCP for impact analysis, tracing, and safe multi-file renames. It suits developers who want their agents to understand architecture. Index your repo and connect your agent.
Project number two, Cursor Plugins, official plugin specification and marketplace for Cursor. Extending Cursor with new tools and integrations lacked a standard structure. This repo defines the plugin format and gathers first-party and third-party plugins, skills, rules, MCP servers into one marketplace, with a scaffolding helper for building your own. It suits Cursor users wanting more capability or a plugin of their own. Browse the marketplace and install one.
Project number three, awesome-gpt-image-2, structured prompt library for GPT-Image2. One-off prompting makes image generation unpredictable and hard to repeat. This project compresses prompts into structured protocols—subject, lighting, layout, materials—backed by a gallery of reverse-engineered examples and an installable agent skill for Claude Code, Codex, and Cursor. It suits designers and developers automating image generation. Browse the cases and adapt a template.
Project number four, Ghidra Software Reverse Engineering Framework. Analyze compiled code across platforms. Reverse engineers often need to understand software without its source. Ghidra, built by the NSA, disassembles, decompiles, and graphs binaries across Windows, macOS, and Linux, running interactively or headless, and is scriptable in Java or Python. It suits security researchers and malware analysts studying binary behavior. Download a release, and open your first binary.
Project number five, TypePHP, ahead-of-time compiler turning PHP into native binaries. Interpreted PHP pays a runtime cost that's hard to escape. TypePHP lowers PHP to C++ 17 and compiles it natively, adding type scalars and containers for hot paths while staying interoperable with the Zend runtime, and can output an executable, extension, or WebAssembly build. It suits PHP developers needing near-native speed. Compile your first PHP file and run it natively.
Project number six, Chrome DevTools for Agents, give coding agents a real browser. Agents debugging web pages usually have to guess at runtime behavior. This MCP server drives Chrome via Puppeteer, letting agents click, navigate, capture performance traces, inspect network requests, and run Lighthouse audits, with a CLI available outside MCP too. It suits developers who want agents to test and debug real pages. Add it to your client, and check a page's performance.
Project number seven, reverse-skill, skill router for security work in coding agents. Agents tackling reverse engineering or pentesting often guess at commands with no consistent method. reverse-skill matches tasks, APKs, binaries, CTF challenges to documented playbooks, checks installed tools, and enforces an authorization gate before acting, logging evidence throughout. It suits security researchers driving work through a coding agent. Clone it, and route your first case.
Project number eight, T3 Code, control your coding agents from anywhere. Running several coding agents means juggling separate interfaces with no remote access. T3 Code sits atop your existing Claude Code, Codex, Cursor, and other subscriptions, giving one mobile, web, and desktop surface to start, monitor, and steer them. Back-end on your own machine, remote-ready. It suits developers running multiple agents who want unified control. Run it and connect your agents.
Project number nine, selfdb, executable programs stored as SQLite databases. Binary formats are opaque to ordinary tooling. SelfDB stores a program's sections, symbols, and dependencies as database rows instead of ELF-F structures, so SQL queries can inspect or even strip a binary that still runs via a custom kernel loader. It suits systems programmers curious about binary formats. Enter the dev shell and run the showcase.
Project number 10, openclaude, one coding-agent terminal for any model provider. Switching model providers usually means switching tools entirely. OpenClaude, built from the Claude Code codebase, gives one terminal workflow across OpenAI compatible APIs, Gemini, local Ollama, and more, with Bash execution, sub-agents, MCP, and a VS Code extension. It suits developers who switch between model providers. Install it and set up your first provider.
Project number 11, Academic Research Skills for Claude Code, full research-to-publication pipeline as skills. Academic writing involves tedious reference hunting and citation checking that eats into research time. This Claude Code skill suite runs literature reviews, drafts and formats papers, and peer reviews them against a 100-point rubric, verifying every citation against Semantic Scholar and Crossref to catch fabrications. It suits researchers writing papers with AI assistance. Install it and plan your paper.
Project number 12, ECC, engineering system layered onto your coding agent. Getting disciplined, verifiable output from a coding agent usually means re-explaining process in every prompt. ECC wraps the agent in a plan-test-implement-review-verify loop, backed by specialized sub-agents, hundreds of skills, runtime hooks, and a shared memory vault, working with Claude Code, Codex, Cursor, and more. It suits developers wanting a disciplined workflow. Install it and plan your next feature.
Project number 13, Crawl4AI, open-source web crawler built for LLMs. Raw web pages are too cluttered with ads and navigation to feed straight to a language model. Crawl4AI fetches pages and returns clean markdown or schema-driven JSON, with full browser control, stealth mode, deep crawling, and MCP integration for tools like Claude Code. It suits developers building data pipelines or knowledge bases for AI. Install it and crawl your first page.
Project number 14, awesome-gpt-image-2, reusable prompt templates for GPT-Image2. Getting consistent results from GPT-Image2 is difficult without a repeatable structure. This library compresses prompts into composable parts—subject, lighting, layout—across categories from UI to product shots, with an installable agent skill and a companion site for browsing and filtering by style. It suits designers and developers automating image generation. Browse the gallery and adapt a template.
Project number 15, chatbot-template, minimal starter for building a chatbot UI. Wiring up a streaming chat interface from scratch wastes time better spent on your actual app. This Next.js and AI SDK starter ships token streaming, tool calling, web search, and human-in-the-loop questions out of the box, deployable to Vercel in one click. It suits developers building a chatbot on this stack. Clone it and start customizing.
Project number 16, claude-mem, persistent memory across sessions for coding agents. Coding agents forget everything once a session ends, forcing constant re-explanation. Claude-Mem captures tool usage, compresses it into semantic summaries in a local SQLite and vector store, and injects relevant memories back automatically via token-efficient MCP search tools. It suits developers who want their agent to retain project knowledge. Install it and let your agent remember.
Project number 17, LoopX, control plane for long-running agent loops. Agent tasks spanning days tend to drift, leaving no clear record of the current goal, or whose decision is authoritative. LoopX sits above runtimes like Codex or Claude Code, holding goals, human gates, and decisions in one reviewable state with a quota guard and dashboard to keep loops on track. It suits people running multi-day agent projects needing human judgment at key points. Install it and connect your first goal.
Project number 18, Praxist, autonomous research loop for measurable projects. When a project's objective is measurable but the best approach is unknown, manual trial and error is slow. Praxist runs parallel research agents across generations, evaluates results into structured evidence, and plans the next round, preserving promising candidates and ranking them on multiple metrics until it converges. It suits teams already iterating on a measurable problem by hand. Point it at your project and start a run.
Project number 19, codeburn, see where your AI coding spend goes. Provider bills show total AI spend, but not what drove it—like an expensive model burning budget on trivial work. CodeBurn reads local session files from tools like Claude Code and Cursor, breaks down cost by task and model, flags waste patterns, and can enforce spending cap hooks. It suits developers wanting honest visibility into AI costs. Run it and see where your budget goes.
Project number 20, Obtainium Emulation Pack, auto-updating emulator collection for Android. Android emulators are scattered across dozens of GitHub repos, making manual updates tedious. This pack bundles source and version detection rules for many console emulators, launchers, and compatibility layers into one importable file, so Obtainium tracks and updates them automatically, with a variant for dual-screen handhelds. It suits Android handheld and retro gaming users. Import the pack and let Obtainium handle updates. Thanks for watching, see you in the next update.
Video
