I'm a solo dev who jumped on the AI coding agent train early (Claude Code, Cursor). For the first few months, I was writing CLAUDE.md files from scratch for every new client repo. Each time: same guardrails, same stack-specific prompts, same debugging of inconsistent agent behavior. It took me about 5 hours per project to get the rules right — and I'd still miss edge cases.
So I finally sat down and extracted the patterns I'd been reinventing: a set of production-tested rule templates for backend (FastAPI, Rails), frontend (Next.js), testing, git workflow, and security. Plus 6 reusable subagent definitions (code-reviewer, tdd-guide, build-error-resolver, security-reviewer, planner) with prompts tuned for those stacks. I also wrote a short guide on wiring them into a new repo in under 30 minutes.
I've packaged it as the "AI Coding Agent Starter Kit" — a downloadable template pack on Gumroad. It's $39 for the core kit, $59 for the Pro tier that adds a Loom walkthrough and 3 stack-specific variants. No fake testimonials, no hype — just the files I wish I'd had 6 months ago.
If you're a freelance dev or small agency hand-rolling agent rules per client, this might save you the same 40+ hours I burned. Happy to answer questions about what's inside or how I structured the subagent prompts.