Original MetaCyberGuru collection

AI code-review prompts for professional developers

These prompts are designed for working engineers. They ask for evidence, version context, failure scenarios and verification. None of them can prove that code is correct. They help you inspect the right things before you make a change.

Give the model enough evidence

Paste the smallest relevant diff or component, not an entire private repository. Include the requirement, runtime, failing test and expected behaviour. Remove credentials, customer data and internal URLs.

Keep the review falsifiable

A useful finding names the exact code, failure path and test that would expose it. Reject vague comments such as “consider improving error handling” unless the reviewer can show what fails.

MetaCyberGuru reviewed
Audit this interface against the supplied interaction and markup. Check semantics, heading order, keyboard operation, focus visibility, accessible names, form errors, contrast assumptions, zoom, motion and screen-reader announcements. Do not infer compliance from a screenshot alone. Mark what needs browser or assistive-technology testing. Markup: [HTML OR COMPONENT]. Styles: [CSS]. Interaction: [USER FLOW].

Customize: HTML OR COMPONENT, CSS, USER FLOW

MetaCyberGuru reviewed
Create a risk-based QA plan for this change. Map each requirement to a test, then cover integration boundaries, permissions, failure recovery, concurrency, data migration, observability and rollback. Prioritize by impact and likelihood. Product change: [CHANGE]. Requirements: [REQUIREMENTS]. Architecture: [ARCHITECTURE]. Known incidents: [INCIDENTS OR NONE].

Customize: CHANGE, REQUIREMENTS, ARCHITECTURE, INCIDENTS OR NONE

MetaCyberGuru reviewed
Evaluate this regular expression as production input validation. Explain what it accepts, then design positive, negative and adversarial test cases. Include empty input, Unicode, extreme length, ambiguous boundaries and likely ReDoS cases. Do not claim the expression is safe unless the evidence supports it. Regex: [REGEX]. Intended rule: [RULE]. Engine: [ENGINE].

Customize: REGEX, RULE, ENGINE

MetaCyberGuru reviewed
Diagnose this bug using only the supplied evidence. Build a timeline, separate observations from hypotheses, rank the likely causes, and propose the cheapest discriminating test for each hypothesis. Do not propose a code change until one cause is supported. Symptom: [SYMPTOM]. Logs: [LOGS]. Recent changes: [CHANGES]. Environment: [ENVIRONMENT]. Reproduction: [STEPS].

Customize: SYMPTOM, LOGS, CHANGES, ENVIRONMENT, STEPS

MetaCyberGuru reviewed
You are reviewing an unfamiliar codebase. Explain the supplied code before proposing changes. Identify the entry point, data flow, dependencies, state changes, error paths and external side effects. Separate facts visible in the code from assumptions. Then list the three highest-risk areas to inspect next. Code: [PASTE CODE]. Runtime and version: [RUNTIME]. Known behaviour: [KNOWN BEHAVIOUR].

Customize: PASTE CODE, RUNTIME, KNOWN BEHAVIOUR

MetaCyberGuru reviewed
Review this change against the stated requirements. Report only issues introduced or exposed by the patch. For each issue, give severity, exact file and line, the failure scenario, and the smallest safe correction. Do not report style preferences unless they cause a defect. Requirements: [REQUIREMENTS]. Patch: [DIFF]. Relevant tests: [TESTS].

Customize: REQUIREMENTS, DIFF, TESTS

MetaCyberGuru reviewed
Review this SQL query using the supplied schema and expected result. Check join cardinality, null handling, duplicate rows, filtering order, transaction assumptions, injection risk and index use. Propose a corrected query only when needed, and explain how to verify it with EXPLAIN or the database equivalent. Database: [DATABASE AND VERSION]. Schema: [SCHEMA]. Query: [QUERY]. Expected result: [EXPECTED RESULT].

Customize: DATABASE AND VERSION, SCHEMA, QUERY, EXPECTED RESULT

MetaCyberGuru reviewed
Convert this feature request into an architecture decision record. State the problem, constraints, quality attributes, two or three viable options, trade-offs, security implications, operational cost and a reversible recommendation. Mark missing evidence as unknown. Feature: [FEATURE]. Existing system: [SYSTEM]. Constraints: [CONSTRAINTS]. Expected scale: [SCALE].

Customize: FEATURE, SYSTEM, CONSTRAINTS, SCALE

How to validate the answer

Run the code or test in the stated environment. For SQL, inspect the schema and an execution plan. For accessibility, combine automated checks with keyboard and screen-reader testing. For security-sensitive findings, use a qualified reviewer and follow your organization’s disclosure process.