Tired of writing commit messages? I built a tool that does it for you.   Introducing git-ai — an AI-powered CLI that lives inside your Git workflow.   What it does:   - Generates commit messages that match YOUR style (learns from your last 20 commits)   - Creates PR descriptions with one command   - AI code review with severity ratings   - Auto-generates changelogs in Keep a Changelog format   - Detects ticket IDs (JIRA, Linear, GitHub) from branch names   - Works with Claude, GPT, Gemini, or fully local with Ollama   Just run:   npm install -g @malikasadjaved/git-ai   git-ai setup   git-ai commit   Three commands. Zero friction. Never write a commit message again.   🌐 Website: https://lnkd.in/dzydYtAj   💻 GitHub: https://lnkd.in/dfEJGaBX   🔗 Connect: https://lnkd.in/dv_dkDsd   Built with frustration from writing commit messages manually. Open source & MIT licensed.   #OpenSource #DeveloperTools #AI #Git #CLI #NodeJS #TypeScript #Programming #SoftwareEngineering

I've been running multiple AI coding agents in parallel on my local setup using git worktree and it's working well so far. At Lago we built a bash script that gives each agent its own isolated environment: dedicated worktree, dedicated port, one command to spin up. Wrote about the setup and the method behind it: https://lnkd.in/dMJZsDsb #AI #Software #GitWorktree #ClaudeCode #AIAgents

Recently, I accidentally fell into a rabbit hole… and instead of climbing out, I opened the source code of an AI coding agent 👀 I’ve been curious about agentic coding for a while, so I gave Opencode a spin. And yes, my selection criteria were extremely rigorous… mainly “ooh nice UI” and “that name sounds cool” 😄 While using it, I noticed something odd (atleast to me): It only listed my sessions when I was inside the same folder where they were created. Switching folders? No sessions. Naturally, curiosity kicked in — so I dug into the source code 🔍 Here’s what I found: 👉 Project Identification Opencode generates a unique project ID to track sessions. It does this in two ways:  - If the folder is a Git repo:      It runs   $ git rev-list --max-parents=0 HEAD       to get the Git directory, then derives the project identity from it.       And here’s the fun part — it caches this inside .git/opencode.   Yes, it casually writes into your .git folder. Pretty clever.  - If Git is not initialized:       It falls back to a constant global value (so basically, all such folders look the same to it, no because there are some other conditions too). 👉 Session Tracking Each session gets its own unique ID. 👉 How it links everything It stores sessions in a SQLite database and connects them using the project ID as a foreign key. You can even find where this DB lives with: opencode db path 💡 Why sessions don’t show across folders? Because each folder = different project ID (especially if Git is initialized). No shared project ID → no shared session list. Honestly, I love these small design decisions. Simple idea, clean implementation, and very “developer-minded.” Diving into source code like this always feels like uncovering tiny engineering stories hidden beneath the UI 🚀 #OpenSource #AI #AgenticAI #CodingAgent #DeveloperLife #SoftwareEngineering #TechDeepDive #CodeReading #LearnInPublic #Git #SQLite #Programming #Developers #TechCuriosity #BuildInPublic #EngineeringInsights #Debugging #DevTools #BackendEngineering #CleanDesign

AI agents don't have Ctrl+Z. When they break your code, they don't undo. They re-read every file, reason about what went wrong, rewrite from scratch, burning tokens on code they already had right. Sometimes the "fix" breaks something else. And the cycle repeats. I built ckpt. It watches in the background and auto-snapshots every step your agent takes. Something breaks? ckpt restore 3. Instant. Zero tokens. The agent can even call it itself. Not a new version control, just a thin layer on top of git. Works with Cursor, Kiro, Claude Code, Codex, Aider, anything. npm install -g @mohshomis/ckpt If you use AI agents for coding daily, try it on your next session and tell me what's missing. github.com/mohshomis/ckpt

#30DaysOfVibeCoding - Day 7: GitDash Same stack as yesterday. Same framework. Not the most original idea either. But this build revealed something worth noting. The AI didn't need to understand git's internals. It just wrapped the CLI. Every piece of data in the dashboard comes from running a git command and parsing the text output: - git status --porcelain for dirty files - git rev-list --count for ahead/behind - git stash list for stash counts It treated git as a black box with a text interface — which is exactly how most developers treat it too. That's the interesting pattern. You can point AI at any CLI tool with structured output and get a wrapper UI built around it. Git, Docker, kubectl, whatever. Is the tool itself useful? Lazygit and tig already exist and do it better. But the capability of generating wrapper UIs for existing CLIs without understanding their internals — that opens a category. GitHub: https://lnkd.in/e6mcTRuM Full write-up: https://lnkd.in/exeeiyqg #VibeCoding #AI #BuildInPublic #Go #TUI #DevTools

Your CLAUDE.md is a suggestion. Your git hooks are the law. We have a monorepo where both engineers and non-engineers build internal tools using Claude Code. We quickly learned that instruction files alone don't cut it. The AI follows them most of the time. But "most of the time" isn't good enough when you have 10+ apps and people who've never written code pushing changes. So we built 9 pre-commit and pre-push hooks that act as a deterministic enforcement layer. The AI's code literally cannot reach the repository without passing them. The three things that surprised us: 1. Hook error messages are prompt engineering. A cryptic "Error: invalid" means the AI spirals. A formatted error with the rule, the violation, and the exact fix command means the AI self-corrects on the first retry. 2. Advisory warnings are useless. The AI ignored every single one. If it's important enough to check, block the commit. 3. The highest-ROI hook we built forces documentation updates. If you change code in an app but don't update the app's CLAUDE.md, the commit is blocked. Documentation stays alive because the system demands it, not because someone remembered. Wrote up the full approach with examples: https://lnkd.in/gxkrQXx8 If your team is using AI coding agents in production, your git hooks are now part of your AI strategy.  #AIEngineering #ClaudeCode #DevTools #SoftwareEngineering

GitHub just shipped an embedding model update that actually matters. Their new Copilot embedding model for VS Code isn't just incrementally better — it's 37.6% more accurate at finding the right code context, runs twice as fast, and uses 8x less memory for indexing. For C# and Java devs, acceptance rates for suggestions have doubled. That's not a feature update. That's a productivity shift. What's interesting: they used contrastive learning techniques (InfoNCE loss + Matryoshka Representation Learning) to train this. The model now powers chat, agent, edit, and ask modes — so the improvements cascade across every interaction you have with Copilot. But here's what I'm thinking about: we're moving from "AI suggests code" to "AI understands your codebase architecture." Better embeddings mean better context retrieval. Better context means suggestions that feel less like autocomplete and more like pair programming with someone who's read your entire repo. For QA folks and builders working in complex codebases, this changes test automation workflows and API integration work significantly. Question: Are you seeing improved Copilot suggestions in your stack after this update, or is acceptance rate still a coin flip for you? #GitHubCopilot #AICoding #DeveloperProductivity #CodeEmbeddings #DevTools

We’ve trusted Git for everything — clean versioning, easy collaboration, and quick rollbacks. But when I started building real ML projects, I realized Git alone wasn’t enough. Git works great for software development, but in ML, data broke everything. Massive datasets, model weights, constantly changing labels, and scattered experiments made versioning a nightmare. Git LFS was expensive, S3 buckets felt disconnected, and reproducibility became painful. That’s when I discovered DagsHub — GitHub for Data Science. It neatly combines Git + DVC + MLflow in one platform. I finally got: - Reliable versioning for large datasets (no more LFS headaches) - Built-in experiment tracking - Free remote storage + model registry I tested it on a project containing audio, images, and tabular data. I ended up tracking 3GB+ of data while keeping my Git repository under 50KB. Clean, reproducible, and actually enjoyable. Want the full story — setup steps, DVC commands, MLflow integration, and key learnings? 👉 Read the complete post here: https://lnkd.in/gdM-ERPk #MLOps #AIOps #DevOps #MachineLearning #ProductionAI #AI

An Anthropic hackathon winner just open-sourced his entire Claude Code setup. It has 117k+ stars on Github -- for teaching Claude how to work better across all these verticals: 1. Agents: Planner, architect, code reviewer, security auditor Build error resolvers across Python, Java, Go, Rust, C++ E2E runners, refactor cleaners, doc updaters Even a “chief-of-staff” agent for communication 2. Skills (100+ workflows) TDD, eval harnesses, verification loops Token optimization + cost-aware LLM pipelines Continuous learning that turns sessions into reusable skills Backend, frontend, DB, DevOps, and even investor workflows 3. Commands (60+) /plan → break down features /tdd → enforce test-first development /verify → run evaluation loops /multi-* → orchestrate multi-agent workflows /pm2 → manage services 4. Rules (always-on constraints) Coding standards, testing requirements, security checks Language-specific best practices (Python, TS, Go, etc.) Enforces consistency across every generation 5. Hooks (automation layer) Save/load memory across sessions Auto-evaluate outputs Suggest compaction before context breaks Trigger logic on every tool call 6. System layer Context injection modes (dev, review, research) MCP integrations (GitHub, Supabase, etc.) Cross-platform scripts + installers Full test suite to validate everything 7. Real examples SaaS apps (Next.js + Stripe) Django APIs Go microservices Rust backends ♻️ Share it with anyone who uses Claude Code :) I share tutorials on how to build + improve AI apps and agents, on my newsletter 𝑨𝑰 𝑨𝒈𝒆𝒏𝒕 𝑬𝒏𝒈𝒊𝒏𝒆𝒆𝒓𝒊𝒏𝒈: https://lnkd.in/gaJTcZBR #AI #AIAgents #LLMs

Anthropic hackathon winners’ setup A fully systematized Claude Code stack—agents, workflows, rules, and automation—turning LLMs from tools into an integrated engineering OS.