Engineering Advanced Vibe CodingAI Pair ProgrammingPrompts

Vibe Coding Guide

Structured AI-pair-programming: prompt decomposition, iterative refinement, context management, quality gates, and session workflows.

A meta-skill for developers who want to code faster with AI without accumulating hidden technical debt. It teaches structured prompt decomposition, context management strategies, quality gates between AI-generated iterations, and the session workflows that keep AI pair-programming sessions focused and productive rather than drifting into spaghetti.

Added Mar 20, 2026

$npx skills add johnefemer/skillfish --skill vibe-coding-guide

Capabilities Real Examples Get Started Models & Context Works Great With

What This Skill Can Do

Concrete capabilities you get when you install this skill.

Decompose complex features into AI-sized implementation tasks with clear interfaces

Design prompt templates that produce consistent, reviewable code increments

Manage context window effectively: what to include, what to summarize, when to reset

Apply quality gates between AI iterations: review checkpoints, test coverage, lint passes

Run structured debugging sessions with AI: hypothesis formation, targeted log instrumentation

Recognize and prevent AI coding anti-patterns: over-engineering, orphaned code, scope creep

Real Conversations

See how the Vibe Coding Guide skill responds to actual requests. These are representative examples of what your AI agent will produce.

You

I want to add Stripe subscription billing to my SaaS. Where do I start with AI-assisted coding? I tend to lose track and end up with messy code.

Agent

The key is decomposing before coding — never start a session with 'add Stripe billing.' That produces a monolithic dump you can't review or test incrementally.

Step 1 — Decompose into vertical slices (5 minutes)

Phase 1: Stripe webhook handler + signature verification
Phase 2: subscription_plans table + DB migrations
Phase 3: checkout session creation endpoint
Phase 4: customer portal endpoint (manage subscription)
Phase 5: subscription status guard middleware
Phase 6: UI: pricing page + upgrade flow

Each phase has a clear interface and can be tested in isolation.

Step 2 — Prompt pattern for each phase

Task: Implement [Phase 1 — Stripe webhook handler]
Context: Express app, TypeScript, Postgres with Drizzle ORM
Constraints:
- Handle checkout.session.completed and customer.subscription.deleted events
- Verify Stripe-Signature header, return 400 on mismatch
- Update users.subscription_status in DB, wrap in transaction
- Return 200 immediately, process async
Do NOT: implement the UI, payment form, or checkout creation (that's Phase 3)
Output: route handler code + the Drizzle DB update function

Step 3 — Quality gate before next phase

Code is under 80 lines
Type errors: zero
Test covers happy path + invalid signature
No TODO comments left in

Don't proceed to Phase 2 until Phase 1 passes the gate. This is the discipline that prevents vibe coding debt.

You

I had a 2-hour AI coding session and now I have 800 lines of code that partially work but have conflicting patterns everywhere. How do I salvage it?

Agent

This is a context drift problem — the AI lost track of your established patterns and started improvising. Salvage in three phases:

Phase 1 — Audit the damage (30 min) Categorize the 800 lines:

- Working + consistent with your patterns → keep
- Working but inconsistent (wrong error handling, import style, etc.) → refactor
- Not working, can be deleted → delete
- Not working, worth fixing → isolate and fix separately

Be ruthless. Code that partially works but is architecturally wrong will cost you more later.

Phase 2 — Extract the good parts Identify the 2-3 functions or modules that are genuinely correct. Extract them into clean files. These become your foundation.

Phase 3 — Restart with constraints For the sections you're rebuilding:

Here's my established pattern from [working file]:
[paste 20-30 lines of canonical code]

Now implement [specific function] following exactly this pattern.
Constraints: same error handling, same import style, same naming conventions.
Output: only the code for this function, nothing else.

Prevention for next time:

Every 45 minutes: review everything generated, run lint + tests
If a session runs >60 lines, start a new chat with a summary of what's been built
Never let the AI touch more than 2 files in one prompt

Get Started

How to install and use this skill in your preferred environment.

Skills are designed for AI coding agents (Claude Code, Cursor, Windsurf) and IDE-based workflows where the agent can read files, run scripts, and act on your codebase.

Models & Context

Which AI models and context windows work best with this skill.

Recommended Models

Works with any frontier model. The skill's value is process and prompt structure, not model-specific capabilities.

Context Window

SKILL.md is ~8KB. This skill is most effective when the user shares their current codebase context and specific feature goals.

Pro tips for best results

1

Be specific

Include numbers — users, budget, RPS — so the skill can size the architecture.

2

Share constraints

Compliance needs, team size, and existing stack all improve the output.

3

Iterate

Start with a high-level design, then ask follow-ups for IaC, cost analysis, or security review.

4

Combine skills

Pair with companion skills below for end-to-end coverage.

Works Great With

These skills complement Vibe Coding Guide for end-to-end coverage. Install them together for better results.

Cursor Rules Engineer

Create optimized .cursorrules and AI editor configuration files from codebase analysis. Boost AI coding assistant accuracy and consistency.

CursorAI Editor

Claude Code Config

Generate optimized CLAUDE.md files, memory configs, and project instructions for Claude Code. Analyzes codebase to create tailored AI context.

Claude CodeCLAUDE.md

Senior Fullstack

Fullstack development with Next.js, FastAPI, MERN, and Django stacks.

FullstackNext.js

TDD Guide

Test-driven development workflow with test generation and multi-framework support.

TDDTesting

$ skillfish add johnefemer/skillfish --all # install all skills at once

Ready to try Vibe Coding Guide?

Install the skill and start getting expert-level guidance in your workflow — any agent, any IDE.