ralphctl v0.5.0 is out 🍩 The headline of this release isn't glamorous, but it matters: the generator-evaluator fix-loop actually works now. When the evaluator flags an issue, ralphctl now properly resumes the generator's session with that feedback instead of starting cold, and skips the redundant final re-eval once the fix lands. There's also a small internal refactor (collapsed option types, shared spawn helper) that makes the evaluate command much easier to reason about. For anyone new here: ralphctl is an agent harness for long-running AI coding tasks. It orchestrates Claude Code and GitHub Copilot across repositories, breaks tickets into dependency-ordered tasks, and runs each one through a generator-evaluator loop so issues get caught before they compound. No flashy new feature this time, but the loop you actually rely on is a lot more trustworthy. See on npm: https://lnkd.in/eYg_GsDs Release notes: https://lnkd.in/eQQ5tJT6 #AgentHarness #ClaudeCode #GitHubCopilot #AugmentedEngineering

We did the math: Our coding agents were working maybe 15 minutes an hour. The other 45 minutes? Waiting on us. That’s when it clicked: the real bottleneck wasn’t the agents. It was the human coordination layer around them. So we built WaveCode. ▸ Claude Code, Codex, and Aider running in parallel ▸ One chat interface across all agents ▸ Agents review each other before code reaches you ▸ Self-hosted, open source, Apache 2.0 The result: 5 agents, 23 minutes, one full feature shipped — backend, API, tests, docs, and a bug caught before the human saw it. The video below shows it running live. GitHub: github.com/dbenic/wavecode More: wavenetic.com/wavecode #OpenSource #AI #DevTools #CodingAgents

Playing around with agents is so much fun! During my (very little) spare time I got Claude to re-write the whole agent client framework for Tinker. One thing I learned more from this re-writing is that agents just don't magicallly invoke the tools (you know, the code for it to interact with the files). The LLM just responds with something like "I want to use tool X for this task" and I have to route the tool X code to it. What seems to be more interesting is that the newer versions of Claude like Opus 4.7 seems to be much eager to use the bash tool. And for Tinker I assign each conversation to a SQLite database because.... it's lightweight I suppose? I was thinking like: What if the agents can just exchange conversations? What if I can summarize the whole conversation, put it in a .db file, and refer to it in another SQLite database? The features are many but the time is just so little. For the next steps: I really want to build a nice sandbox environment, then clone Tinker's source code there, then the agent can finally build itself (with restrictions). Finally, a self-driving code base! And maybe moving the entire agent platform to Kubernetes, you know, for fun? :) There are just so many possibilities. P/S: The Discord client with the thread feature is really nice as well. #agent #stripeminions #rampinspect #ai #llm #coding

🌟 New GitHub Discoveries — 10 fresh repos 1. n8n-io/n8n ⭐ 184,393 stars Fair-code workflow automation platform with native AI capabilities. Combine visu https://lnkd.in/dgNJVu3Q 2. ollama/ollama ⭐ 169,218 stars Get up and running with Kimi-K2.5, GLM-5, MiniMax, DeepSeek, gpt-oss, Qwen, Gemm https://lnkd.in/dKPdRGVt 3. affaan-m/everything-claude-code ⭐ 159,005 stars The agent harness performance optimization system. Skills, instincts, memory, se https://lnkd.in/dXqu9W34 4. anomalyco/opencode ⭐ 144,642 stars The open source coding agent. https://lnkd.in/d5AQXYHy 5. langchain-ai/langchain ⭐ 133,830 stars The agent engineering platform https://lnkd.in/dYV8M_hK #GitHub #OpenSource #Developer #Tech

🌟 New GitHub Discoveries — 10 fresh repos 1. n8n-io/n8n ⭐ 184,407 stars Fair-code workflow automation platform with native AI capabilities. Combine visu https://lnkd.in/dgNJVu3Q 2. ollama/ollama ⭐ 169,229 stars Get up and running with Kimi-K2.5, GLM-5, MiniMax, DeepSeek, gpt-oss, Qwen, Gemm https://lnkd.in/dKPdRGVt 3. affaan-m/everything-claude-code ⭐ 159,193 stars The agent harness performance optimization system. Skills, instincts, memory, se https://lnkd.in/dXqu9W34 4. anomalyco/opencode ⭐ 144,737 stars The open source coding agent. https://lnkd.in/d5AQXYHy 5. langchain-ai/langchain ⭐ 133,851 stars The agent engineering platform https://lnkd.in/dYV8M_hK #GitHub #OpenSource #Developer #Tech

704 tokens explain bug. The fix was three lines. The other 700 tokens? "Sure! I'd be happy to help you with that!" A developer named Julius Brussee got tired of paying for AI pleasantries. He built Caveman. It is a Claude Code skill that makes your AI agent talk like a caveman. Not dumb. Efficient. The results: 87% token savings on React debugging 83% on auth token fixes 81% on PostgreSQL race conditions 65% average across all tasks But here is the part nobody expected: A March 2026 paper tested 31 models on 1,485 problems and found that brevity constraints improved accuracy by 26 percentage points. Big models talk much. They reason themselves into wrong answers. Caveman stops overthinking. One install. Zero dependencies. MIT license. Works on Claude Code, Codex, Gemini CLI, Cursor, Copilot, and 40+ more agents. Caveman not dumb. Caveman efficient. Caveman say what need saying. Then stop. Full breakdown in the carousel and blog post at imiel dot dev. #AI #ContextEngineering #DeveloperTools #ClaudeCode #Caveman #TokenOptimization #OpenSource

7% token savings on React debugging. Not from a new model. Not from a better prompt. From telling the AI to stop being polite. A developer built a Claude Code skill called Caveman that strips the pleasantries out of AI responses. No "Sure! I'd be happy to help you with that!" Just the fix. Three lines instead of 704 tokens. The part that matters for anyone running AI in production: a March 2026 paper found that brevity constraints actually improved accuracy by 26 percentage points across 31 models. The AI wasn't just wasting tokens on small talk. It was reasoning itself into wrong answers. One skill file. Zero dependencies. Works across Claude Code, Cursor, Codex, Gemini CLI, and 40+ agents. This is context engineering in its simplest form. Tell the tool what not to do. Full breakdown in the original post from Imiël Visser.

The AI dev stack got fast at writing code. It didn't get any better at showing you what that code does after deploy. New post: how we merged Vercel + Supabase logs into a single Gonzo terminal session for real-time cross-platform debugging. No log drains, no third-party platform, four lines of bash. → https://lnkd.in/gi_4V_hz #vibecoding #observability #devtools #opensource

🔥 I built an AI Agent that reads, understands, and debugs entire GitHub repositories. Not just file-by-file. The whole codebase , at once. After days of deep building, I'm launching GitHub Smart Agent, a production-ready AI tool that turns any GitHub repo into an interactive, intelligent assistant. What it actually does: 🔍 Ingests any GitHub repository in seconds 💬 Chats with your entire codebase like a senior dev 🧠 Detects bugs using both static analysis + AI reasoning ⚡ Suggests optimizations with context-aware explanations 📊 Generates structured analysis reports — ready to share Under the hood: → LangChain + FAISS Vector Store for RAG pipeline → OpenRouter (Gemini Flash / DeepSeek) as the reasoning engine → Custom embedding pipeline for code-aware retrieval → Streamlit with a fully custom premium UI Why I built this: Every developer has opened a legacy codebase and thought "where do I even start?" This tool answers that question in seconds. 🎬 Full demo video is live — watch it below. 🔗 GitHub repo dropping in the next day or two — I'll post when it's up. Drop a follow so you don't miss it! If you're working in AI engineering, LangChain, LLM tooling, or developer tools , I'd love to connect and exchange ideas. What's the biggest pain point you face when navigating an unfamiliar codebase? 👇 #AI #LangChain #RAG #LLM #DeveloperTools #Python #Streamlit #CodeAnalysis #OpenRouter #FAISS

Stop letting flaky LLM calls crash your AI Agents. 🛡️ I’m excited to announce the release of Veridian Guard (v0.2.2) — a lightweight, zero-dependency safety layer for Python-based autonomous systems. While building AI agents, I noticed a recurring pain point: LLM APIs are inherently unpredictable. A single timeout or rate-limit shouldn't bring down an entire production workflow. That’s why I built Veridian Guard. It’s designed for developers who value system reliability and clean, maintainable code. Key Features: 🛡️ The @guard Decorator: One line of code to protect any function. ⚡ Smart Auto-Detection: Seamlessly handles both synchronous (def) and asynchronous (async def) functions out of the box. 🪵 Built-in Observability: Smart logging to help you identify exactly where and why your agent struggles. 📦 Pure Python: Zero external dependencies. Fast, secure, and lightweight. In an era of autonomous systems, engineering for "error tolerance" isn't just an option—it's a requirement for building scalable and trustworthy AI ecosystems. I’d love for you to try it out, star the repo, and share your feedback! 📦 PyPI: pip install veridian-guard 💻 GitHub: https://lnkd.in/erD78Bvi Happy coding! 🌿 #Python #OpenSource #AIAgents #SoftwareEngineering #LLM #Reliability #AutomotiveSoftware #TechInnovation