Klerith

Klerith / fernando-skills

Public

Spec-driven development skills for Claude Code, Cursor, Codex, and Antigravity. By Fernando Herrera.

28
0
89% credibility
Found May 19, 2026 at 28 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Shell
AI Summary

This is a collection of two AI assistant skills that introduce a structured way to build software features. Instead of jumping straight into coding, the skills guide you through creating a detailed specification document first—where you describe the problem, the AI asks clarifying questions, and together you define exactly what will be built and how. After you manually review and approve the specification, a second skill implements it one small step at a time, pausing after each step so you can review the changes before continuing. The method is designed to prevent AI assistants from making invisible decisions that become expensive to undo later, and to keep humans in control of important design choices.

How It Works

1
💡 You discover a better way to work with AI

You've been frustrated watching AI assistants make dozens of decisions without you noticing, leading to code that's hard to change later.

2
📦 You add the skills to your AI assistant

With one simple command, you install two new abilities into your coding assistant that change how features get built.

3
🎯 You describe what you want to build

Instead of telling the AI how to build it, you explain the problem you're solving and let it ask you questions to understand your needs.

4
📝 The AI creates a detailed plan for your review

Your AI assistant writes out a complete specification: what will be built, what won't be built, how the data will work, and how it will be built step by step.

5
✋ You read it carefully and give your approval

You step away from the chat, read the plan on your own, make any changes you want, and manually change its status to 'Approved'—only you can do this.

6
The AI builds your feature piece by piece
✅
Everything looks good

You approve the changes and the AI continues to the next step

🔄
Something needs adjusting

You catch an issue early while it's still easy to fix, before more code is written on top of it

🎉 Your feature is built exactly as you planned

Every decision was made intentionally, documented, and reviewed by you. The code matches the plan, and you understand exactly what was built and why.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 28 to 28 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is fernando-skills?

This is a collection of AI coding skills for Claude Code, Cursor, Codex, and Antigravity that enforce a spec-driven workflow. Instead of letting an AI agent barrel ahead and make implicit design decisions, you write a detailed specification first, approve it manually, and then the agent implements it step by step with pauses for review. The two core skills are `/spec` (which designs a feature spec through clarifying questions) and `/spec-impl` (which implements an approved spec on a dedicated branch). Everything lives in markdown files versioned in git, forming a living design decision log for your project.

Why is it gaining traction?

The hook is straightforward: AI coding agents are fast but reckless. They make 50 implicit decisions in 30 seconds that cost you hours to undo later. This project forces a human checkpoint between planning and implementation. You describe the problem, Claude proposes a structure, you iterate with concrete decisions, then manually flip the spec to "Approved" before the agent can touch any code. The step-by-step execution with built-in pauses means you review 50-line diffs instead of 600-line dumps. It's a lightweight process framework that fits in a single README but addresses a real pain point developers experience with AI agents going off the rails.

Who should use this?

Developers who use AI coding agents and have felt that frustration of watching an agent generate clever but wrong code you now have to untangle. It's particularly useful for features spanning multiple files or sessions, where design decisions need to persist and be referenced later. If you're building something with non-trivial data schemas, APIs, or state management, this gives you a structured way to lock those decisions in before implementation starts. Not for quick bug fixes or one-off refactors.

Verdict

The concept is solid and the documentation is thorough for a 28-star project. At this credibility score and star count, treat it as a promising early-stage experiment from a known educator rather than a battle-tested framework. If you're already using Claude Code or Cursor and want more control over AI-generated code, the workflow is worth trying. Start with a small feature, see if the spec discipline catches real problems, and decide from there.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.