Developers use ChatGPT for speed, but unstructured requests produce code that compiles nowhere and advice that ignores your stack. These prompts add repository context, constraints, and verification steps so outputs are mergeable—not museum pieces.

Auto-build stack-aware prompts with Generate a prompt. For Claude-specific review rubrics, see Claude Prompts for Code Review. Upload logs or specs via upload when you need grounded answers.

Rules for every developer prompt

Name language, framework, and version (e.g. TypeScript 5.4, Next.js 15 App Router)
Paste minimal reproducible snippet, not whole repos
State what you already tried
Ask for tests or types when production-bound
Request “explain assumptions” before large diffs

Debugging

Root-cause analysis

You are a senior [LANGUAGE] debugger.

**Symptom:** [ERROR MESSAGE / BEHAVIOR] **Environment:** [OS, RUNTIME, FRAMEWORK VERSIONS] **Relevant code:** [PASTE]

List top 3 hypotheses ranked by likelihood. For each: what evidence supports it, specific next experiment (log line, breakpoint, curl), and fix if confirmed. Do not propose rewrites unrelated to symptom.

Minimal repro extraction

From this large file, extract the smallest snippet that could reproduce [BUG] without proprietary names. Replace secrets with placeholders.

Log interpreter

Parse these logs chronologically: [PASTE]. Timeline of events, first failure point, correlated warnings, suggested metrics to add. No code changes yet.

Implementation

Feature with tests

Implement [FEATURE] in [LANGUAGE/FRAMEWORK].

**Requirements:** [BULLETS] **Files:** [PATHS] **Constraints:** match existing style [PASTE SMALL EXAMPLE], no new dependencies unless justified.

Deliver: code diff, unit tests, migration note if DB touched.

API endpoint design

Design REST endpoint for [RESOURCE]. Methods, status codes, request/validation schema, error shape, idempotency for [OPERATION], OpenAPI YAML fragment.

SQL query

Write [Postgres/MySQL] query for [GOAL]. Tables: [SCHEMA]. Explain index usage. Warn about N+1 if ORM example included.

Regex with explanation

Create regex for [PATTERN]. Provide test strings pass/fail, explanation of groups, and safer non-regex alternative if applicable.

Refactoring

Safe refactor plan

Refactor [MODULE] to [GOAL: extract service, remove duplication]. Steps ordered for small PRs. Risks, test coverage needed, rollback strategy. No code until I approve plan—then step 1 only.

Performance pass

Profile mentally this hot path: [PASTE]. Identify allocations, awaits, query count. Suggest fixes with expected impact (high/med/low). Prefer measurement hooks over premature optimization.

Type tightening

Add strict types to [FILE]. No `any`. Preserve runtime behavior. Note breaking type changes for callers.

Testing

Unit tests from spec

Write tests for [FUNCTION] using [JEST/VITEST/PYTEST]. Cases: happy path, edge [LIST], error [LIST]. Mock only [DEPENDENCY].

flaky test diagnosis

Test flakes intermittently: [PASTE TEST]. List timing/async/shared-state causes and stabilization pattern without weakening assertions.

Documentation

README section

Write README section: install, env vars table, dev server, test command, deploy overview for [PROJECT TYPE]. Commands must match [PACKAGE MANAGER].

Docstring pass

Add docstrings to public functions in [PASTE]. Google style / NumPy style [PICK]. Include raises and examples for non-obvious params.

Changelog entry

From commits summary: [PASTE]. Conventional changelog section: Added/Changed/Fixed, user-facing tone, breaking callouts.

Code review (ChatGPT)

Review this PR diff as staff engineer. Tags: BLOCKER/MAJOR/MINOR. File:line references. Merge recommendation. Do not invent files.

Cross-check patterns in Claude Prompts for Code Review—many transfer directly.

Architecture

ADR draft

Context: [PROBLEM]. Options A/B/C with pros/cons. Recommend [OPTION] for team size [N] and traffic [SCALE]. Consequences and migration steps.

Microservice boundary check

Given modules: [DESCRIPTION]. Suggest service boundaries or monolith reasons. Data ownership and sync risks.

DevOps and infra

Dockerfile slimming

Review Dockerfile: [PASTE]. Multi-stage build, layer cache, non-root user, healthcheck. Pin base image tags.

GitHub Actions workflow

Workflow for [CI TASK] on [STACK]. Cache deps, parallel jobs, secrets via GitHub secrets names only, fail fast on lint.

Incident postmortem draft

Timeline: [PASTE]. Blameless postmortem sections: impact, root cause, contributing factors, action items with owners. No individual blame.

Security

Threat model light

Feature: [DESCRIPTION]. STRIDE table for components [LIST]. Top 3 mitigations with code-level hooks.

Secret scan response

I accidentally committed [SECRET TYPE]. Steps: rotate, purge history cautiously, prevent recurrence tools. Do not repeat secret in output.

Data and prompts/meta

Prompt for your own CLI tool

Generate system prompt for CLI that [TASK]. Include argument schema, error messages, exit codes.

Meta: What is Prompt Engineering? A Beginner's Guide.

Frontend

React component

Component [NAME]: props TypeScript interface, accessibility (keyboard, aria), loading/error states, Tailwind classes matching [DESIGN TOKENS].

CSS layout debug

Layout broken at [BREAKPOINT]. HTML/CSS: [PASTE]. Explain box model issue and minimal fix.

Learning and explanation

Explain like senior review

Explain [CONCEPT] to engineer who knows [PRIOR KNOWLEDGE]. Analogy, pitfalls, one production example, links to official docs topic names only.

Compare two approaches

Compare [A] vs [B] for [USE CASE] at scale [TRAFFIC]. Decision matrix: complexity, ops burden, cost, team skill fit.

When ChatGPT struggles

Huge files: chunk and summarize first
Missing private API docs: paste official excerpt
Hallucinated packages: ask “verify package exists on npm/PyPI”

Switch models or add Role Assignment Technique in AI Prompts for architect vs. implementer splits.

Team practices

Store vetted prompts in repo `docs/ai-prompts.md`
Require tests in same PR as AI-generated core logic
Track API spend: pricing on prmptly.dev for teams

Summary

Treat ChatGPT like a junior with infinite stamina: precise context, explicit constraints, verification steps, and small diffs. The prompts above are starting points—encode your stack’s non-negotiables in bracket variables, review like any human contribution, and keep a library of prompts that survived production.