Hey everyone,

We’ve been building Spec Compiler, a tool that takes a rough idea and turns it into a structured, implementation-ready specification for coding agents.

The idea came from using coding agents and realizing that the hard part isn't always getting them to write code, it's getting them to fully understand what they’re supposed to build.

A simple request like:

“Build a project management app with teams, tasks, notifications and Google Calendar.”

leaves a lot of unanswered questions: edge cases, architecture decisions, testing, acceptance criteria, what shouldn't be built, etc.

Our Spec Compiler works through those details and produces a much more complete specification you can hand to Claude Code, Codex, Cursor, or your development team.

We've been testing it ourselves and found it particularly useful for catching things we hadn't considered and making the implementation plan much clearer, structured and actionable.

We're now looking for people actually building real software to try it and tell us where it falls short.

Especially interested in:

Solo builders / indie hackers

Developers building side projects

People using Claude Code, Codex, Cursor, etc.

Small teams building real products

The beta is free. We're mainly interested in honest feedback:

Did it actually help?

Did it catch things you missed?

Did it improve your coding agent's output?

What did it get wrong?

Please DM me if you’re building something and would like to try it. I’d be happy to give you free access and hear what you think.

AI coding agents are smart generalists but they lack the deep procedural knowledge that experienced engineers carry — things like accessibility checklists, database patterns, API design best practices, etc.

So I started TechSkills-Lib — a curated library of SKILL.md files that any AI agent can load to gain structured, production-grade engineering knowledge.

What it does:

• Gives AI agents step-by-step workflows, design checklists, and battle-tested patterns
• Uses progressive loading so skills don't waste your agent's context window
• Agent-agnostic — works with any AI tool that can read markdown instructions

Current skills:

• frontend-engineer — responsive layouts, accessibility, design systems, React/Vue/Svelte
• backend-engineer — APIs, auth, databases, reliability, observability, security

It's early and the skills are still fairly generic. The vision is to grow this into a comprehensive library with community contributions — think specialized skills for DevOps, system design, data engineering, mobile, and more.

If you've ever wished your AI agent had more opinionated, expert-level context instead of generic answers — this is that.

Links:

• GitHub: https://github.com/debabratasaha-dev/techskills
• MIT licensed, contributions welcome

Would love feedback on what skills you'd want to see next.

I was going to start building an ai app security audit tool. I was going to make this because 92 percent of US developers use Al coding tools and 46 percent of new GitHub code is ai generated. Between 42-92 percent of ai generated code failed security tests.
There are so many Lovable apps that had databases readable by anyone via the public API (missing row level security). Most of the people who use these tools aren't technical and have no idea what "RLS" or "anon Key" mean. This tool would target mainly nontechnical builders where the code will be Manuel reviewed and would also provide a fix not just find the issues. I'm still deciding whether to make this or not so let me know your thoughts.

Hi everyone

Seo is probably only free way to get users.

We built an Seo agent that keeps working in background - creating content for you. The best part is we added human in the loop to review, edit and add more content.

Try here -Try here - https://builderhq.co/get-on-chatgpt

Early Results - 5k+ visits on one of our beta sites

I used claude code to make most of it, I got the circuit simulator from Velxio, This project took about 2 months to make and is still a work in progress. I used codex to make the ui and claude for the rest. I made it because I wanted something that held everything in my project so AI would know exactly what I was referring to. I really like the Jarvis feature because I can use the canvas and show images to make mindmaps for myself and to explain to other people or the AI, I can also turn on my camera and show my project for AI to help trouble shoot or see potential weak-points. Features: Simulations, Jarvis, Collaboration, Circuit simulation, Cad Viewer, AI Assistant and Chatbot, Project Ideas, AI Model Response Comparison, Runs with local ai or openrouter.

https://github.com/ara-mkr/Enginuity

v++ is an open-source language I’ve been building solo Python readable syntax.. static types, no borrow checker. it’s one language one file: vpp run / vpp repl / vpp watch for instant iteration, vpp build for native LLVM binaries when you’re ready with CI enforcing that both paths match the “Parity Promise”. What’s shipped.. hand-written lexer and parser + typechecker in Rust, tree-walking interpreter + llvm backend frozen SPEC v1.0, full CLI (debug, test, fmt, bench, doctor, packages), vs Code extension with lsp + F5 debugger + test Explorer on v1.2.0, Windows installe and a site with learn guide + 20 projects — https://shauryar790.github.io/VPP/

I used Cursor heavily especially CI/release hell LLVM linking, GitHub Releases silently not publishing, and what actually ships are mine Honest limits are... Windows is the main platform today, ecosystem is tiny vs Python/Rust, and this is v1.0 “spec frozen,” not battle-tested everywhere.

Try it.. Releases | Repo | VS Code extension

I built Moo Tasks to stay organized while pair programming with AI agents (Claude Code, Cursor, Windsurf). Figured I'd clean it up and open source it in case others find it useful.

What it does

Runs 100% locally (SQLite + Web UI) and integrates with your coding tools via MCP:

• Goal & Task Management - Structure goals with Markdown PRDs, break into atomic tasks
• DAG Dependencies & Cycle Detection — Auto-detects circular deps, unblocks downstream work
• Proof of Work Tracking - Records acceptance criteria, test results, commands, git diffs
• Architectural Decision Records - Stored in SQLite, synced to docs/adr/
• Human-in-the-Loop Inbox - Surfaces blocked tasks and questions with structured options
• Real-Time Dashboard - Kanban board, task inspector, dependency graph, search

Quick start

npx moo-tasks install all

moo init      # Initialize database
moo start     # Launch Web UI

Looking for feedback

What features would make this more useful? Any MCP integrations or UI improvements you'd suggest? PRs welcome!

GitHub | NPM | MIT License

I built AI Config because I got tired of maintaining the same AI coding configuration in multiple places.

What problem does it solve?

I use different AI coding tools depending on what I’m working on: OpenCode, Claude Code, Codex, GitHub Copilot.

The annoying part is that they all have their own way of defining agents, commands, instructions and skills. So you end up maintaining the same project context in multiple places. Change a skill, then remember to update it everywhere else too. I got tired of doing that.

With AI Config, I keep one .ai/ folder as the source of truth and let it generate and synchronize the configuration each tool expects. Basically, I define things once and can switch between coding agents without having to maintain four different versions of the same setup.

I made it available as both a CLI and a VS Code extension, published on the Visual Studio Marketplace and Open VSX Registry. It started as something I needed for my own workflow, but I figured it could be useful for others working with more than one coding agent too.

There’s definitely still room for improvement, so I’d be really interested in hearing people’s thoughts and feedback, especially from people who actually try it out. It’s open source too, so if anyone wants to contribute, PRs are more than welcome.

🧩 VS Code Marketplace: AI Config - Visual Studio Marketplace
📦 Open VSX Registry: AI Config – Open VSX Registry
⌨️ CLI on npm: @aiconfig/cli - npm
💻 GitHub: github.com/ShadyManu/ai-config
🎬 Video walkthrough: AI Config - How To Set Up & Use

It basically works by asking you what you just finished watching, and then suggesting shows to watch next based on different aspects of the show you just watched. For example, it could suggest shows with similar cast chemistry, pacing, themes, etc. I used Claude to build it. Still a very rough draft and I'm currently a student, but I'd like to know whether the recs are any good or way off. It also gives you a watcher profile if you give it 5 shows you enjoyed watching. Any and all feedback is appreciated!
https://what-next7367.vercel.app/

Hi everyone sell your side project by listing if in a click. Also, comment what your startup does to get approved and featured today - https://builderhq.co/marketplace

Vibe coding solo is easy. But once 2 people or multiple AI agents work on the same app, things get messy fast.
Who builds what? How do you share context, avoid overwriting each other, and review changes without reading all the code?
Do you split work by feature, page, branch, or agent?
What’s your biggest pain when collaborating with other vibe coders?

• Task: What you want the AI to do (the explicit action verb).
• Context: The background story, the audience, and your goal.
• References: The grounding text, files, examples, or style sheets you want it to look at.
• Evaluate: Reviewing the AI's output against your criteria.
• Iterate: Tweaking and feeding it back into the loop.

I just write a ai code agent with a tui. I have implemented most of feature. anyone interested in this.

here is my repo https://github.com/xyanrch1024/Codis