Free, copy-ready prompts for code review, debugging, refactoring, test generation and docs — tuned for ChatGPT, Claude, GitHub Copilot Chat and Cursor. Click any card to copy.
Replace [bracketed] variables with your specifics. Each template names the stack, scope and output shape — the three dimensions that drive code-prompt quality.
Tight scope; rationale-per-change.
You are a senior [language] reviewer. Review the function below for runtime errors, type-safety issues, and framework anti-patterns. Return a corrected version + a 1-line rationale per change. No style nitpicks. Stack: [stack].
Ranked hypotheses prevent confident-but-wrong fixes.
You are debugging a [stack] issue. Given the error and code below, produce: (1) most likely root cause, (2) 2-3 alternative hypotheses ranked by likelihood, (3) minimal repro, (4) suggested fix with diff. Do not speculate beyond the evidence.
Behavior checklist catches drift.
Refactor the function below to [goal: extract pure logic / improve readability / reduce branches]. Preserve behavior exactly. Return: refactored code + a 1-paragraph rationale + a checklist of behaviors preserved.
AAA + edge case checklist.
Write unit tests for the function below using [test framework]. Cover: happy path, edge cases (empty, null, max), error cases. Each test ≤10 lines. Use AAA structure. No flaky timing-dependent tests.
Behavior-diff comments make review easy.
Migrate the file below from [old framework/library] to [new framework/library]. Preserve all public types and tests. Flag any behavior change that is not 1:1 with a comment // BEHAVIOR CHANGE.
Type info + realistic example.
Write JSDoc/TSDoc for the function below. Include: 1-line summary, @param with types and examples, @returns with shape, @example with realistic usage, and @throws if applicable. No filler.
For PR review of system-level changes.
Review the architecture diff below. Surface: (1) breaking changes, (2) hidden coupling, (3) failure modes not in the diff, (4) migration risk. Skip naming nits. Return as a markdown table.
Ranked impact — prioritized fixes.
You are a senior [language] performance engineer. Audit the code below for hot-path inefficiencies (allocations, N+1 queries, unnecessary re-renders, IO in loops). Return a ranked list of optimizations with: expected impact, change required, and one-line code sketch. Stack: [stack].
For piping into the next agent step.
Given the user request below, output a JSON plan for an agent with keys: steps (ordered list), each step having: {tool, args, success_criterion}. Tools available: [tool list]. Return strictly valid JSON, no prose.Severity-ranked; actionable.
You are a security reviewer. Audit the code below for: injection vulnerabilities, secret leakage, auth bypass, unsafe deserialization, SSRF, CSRF. For each finding: severity (P0/P1/P2), 1-line description, 1-line fix. No theoretical risks — only exploitable ones in this code.
Specify language, framework, version. "Next.js 15 App Router, React 19" lands very differently from "React".
"Return: corrected code + 1-line rationale per change" beats "explain the code" by a wide margin.
For debug prompts, narrow to one file or one function. Wide scope produces vague hypotheses.
Always paste the actual code. Never make the model infer it from prose.
"Preserve behavior exactly", "no naming changes", "no new dependencies" — explicit guardrails prevent unrequested rewrites.
For agent workflows: "Return as JSON with keys: file, change, rationale". Free-form output breaks pipelines.
Use the AI Prompt Fixer VS Code & Cursor extension to optimize prompts right from your editor — or try the live demo on any prompt above.