GitHub - bewithkhan/bansat: Bansat (بنسټ) — lay the foundation before you build. A blueprint-first framework and Claude Agent Skill by Muhammad Tufail Khan

Why it matters

Muhammad Tufail Khan has released Bansat, an open-source framework distributed through GitHub that forces AI coding agents to plan before they build. It is offered as an Agent Skill for Claude, and its central idea is simple: no production code is written until a human approves a blueprint covering product scope, architecture, security, privacy law, accessibility and every UI state. The motivation is a familiar pattern in agent-built apps, where the demo works but decisions nobody made slip through: Row Level Security left disabled, fonts loaded from Google leaking visitor IP addresses before consent, and Hotjar or Clarity recording sessions without permission. Bansat formalises approval gates: one before code, one before any real personal data is collected, and one before launch, with blocking checks that must pass first. A retrofit track audits already-built apps, reverse-engineers a blueprint from the code and turns each gap into a scheduled fix. The framework is written as a single Markdown document that agents read selectively, making it usable across other tools that accept rules files.