AI Developer Toolkit

Get thorough code reviews with actionable feedback tailored to your language, framework, and standards.

Numbered criteria give both models a checklist to follow. Asking for the problematic line + fix prevents vague feedback.

Referencing CWE numbers gets ChatGPT to think in terms of specific vulnerability classes rather than vague 'security issues'. The attack scenario requirement proves the vulnerability is real.

The frequency context prevents ChatGPT from micro-optimizing cold code paths. 'Only flag issues that matter at the stated scale' keeps recommendations practical.

Including what you've already tried prevents both models from suggesting obvious fixes. Asking to rank by likelihood gets you the most probable answer first.

The ranked probability approach prevents ChatGPT from going down rabbit holes. Including what you've tried avoids circular suggestions.

'Identify the line in MY code' focuses ChatGPT on actionable lines rather than explaining library internals. The root cause vs. symptom distinction prevents superficial fixes.

Constraints force ChatGPT to evaluate options against your real-world limits, not theoretical ideals. The review date ensures decisions don't become permanent by default.

Stating query patterns upfront produces correctly indexed schemas. The 'one thing to change at scale' question surfaces design decisions you'll need to revisit later.

Domain context helps both models give relevant advice. The evaluation checklist ensures comprehensive coverage rather than surface-level review.

The 'why this case matters' comment forces ChatGPT to justify each test, filtering out redundant cases. The 'no mocks unless external' rule produces tests that actually test behavior.

'Realistic data, not placeholder strings' produces documentation that developers can actually test with. The curl example makes the endpoint immediately testable.