Hey everyone. I built SpecRabbit to solve a problem I kept hitting personally: AI fills in architectural gaps you never specified, and you spend more time fixing wrong decisions than you saved.
Curious what workflows you're currently using before you start generating code with Cursor, Claude Code, or similar tools. Text descriptions? Markdown specs? Just winging it and fixing after?
SpecRabbit replaces text-based architecture specs with a visual typed node graph. Unlike prose descriptions, it enforces completeness - you can't add an endpoint without auth level or a database connection without query to run. The export is raw JSON/YAML graph topology with built-in AI instructions, not a screenshot or markdown. Every node combines structured parameters with free text for behavioral context. The result: AI coding agents generate code matching your design, not their assumptions.
