I am gonna be real, AI agents are amazing, but they can also be absolute chaos. I can't count the number of times I asked an AI agent to fix a tiny alignment issue, only for it to rewrite three core components or leave me with a bunch of lint errors. If you've been using tools like Claude Code, Copilot, Gemini, or Antigravity, you've definitely experienced this where the agent just generates code blindly without following any strict rules or architecture.
To protect my own sanity (and my codebase), I built Brokeflow. It's a completely open-source, zero-subscription tool designed specifically for student developers who want to set up strict guardrails and constraints for their AI coding assistants.
Putting an End to Mindless Code Generation
The main idea behind Brokeflow is to make specs-driven development easy since it's kinda new for student developers and it's harsh to just navigate and copy-pasting rules around the web. Instead of writing custom instruction files from scratch, you just answer a few simple questions about your project tech stack, directory structure, styling guidelines, and preferred libraries.
Straight to the game, Brokeflow then generates tailored, ready-to-use configuration files like AGENTS.md, .claudecoderc, or system prompt constraints. You drop these right into your repository, and the next time your AI agent boots up, it reads these instructions first. It instantly knows your rules: don't use Tailwind unless asked, always run npm run build before declaring victory, keep files under 300 lines, etc.
Why It Matters
By establishing these boundaries, it stops the agent from going rogue. It keeps the AI aligned with best practices and your project's unique quirks. Building this was an absolute eye-opener for me on how much better agents perform when they are given structured, explicit boundaries rather than just general instructions.