Cursor

Markdown files for code review: they surface feedback on the code you ask to review and do not modify the codebase unless you explicitly ask for that. The goal is to give you structured input so you can make the final decision.

Prompt packs grouped by concern (foundation, API, database, features, production readiness) with unified shapes (objectives, scope, decision rules, severity rubrics, and review output) so runs stay comparable across services and repositories. They are aimed primarily at the Cloudflare suite alongside TypeScript, with examples and framing in prose.

The repository stays Markdown only: files live under an ai-testers tree; there is no application code here beyond the prompts themselves.

Organised libraries under the ai-testers tree, split by theme (readability, API and database groups, feature-focused reviews, production-readiness gates). I started this when I wanted fast sanity checks while coding; it is now fully integrated into how I work. The set is ongoing and changes often as I refine prompts.

I author and curate the prompt sets, iterating on structure so each file is self-contained enough to run in isolation yet consistent enough to chain as a pipeline when I want a fuller pass.

I treated the work like internal tooling: tightening scope per group, cross-linking related concerns (security with data access, observability with rollout risk), and keeping severity language explicit so review output stays actionable rather than prescriptive.

Migrate and fork prompts into language-specific packs (for example Python, Java, or PHP) aligned with whatever stack I work with next, while keeping the same review discipline.