vaibhav-space/my-ai-coding-workflow
THREE-STAGE AI CODING PIPELINE
A set of instructions and skills for production work with AI agents. Includes a three-stage review pipeline (planning → critique → diff audit) built on 4500+ hours of experience.
// KEY FEATURES
- Three-stage review: planning, plan critique, diff audit
- Curated skills: Superpowers, Ponytail, I Have ADHD
- Principles: correctness over speed, simplicity, verification via diffs
- Agent-agnostic: works with Claude Code, Cursor, Gemini CLI, etc.
// DEEP DIVE: HOW IT WORKS
The repository contains engineering instructions (CLAUDE.md), curated skills (skills/), and a three-stage review pipeline. The agent first writes a plan after exploring the codebase, then an independent critic prompt reviews the plan in a fresh context, and after implementation the developer audits git diff before committing. All stages are captured as prompts for repeatability.
Unlike prompt collections, this is a battle-tested setup refined over 4500+ hours across Cursor, Claude Code CLI, Antigravity IDE, Gemini CLI, Codex, OpenCode, Amazon Q. Includes attributed skills: Superpowers (Jesse Vincent), Ponytail (Dietrich Gebert), I Have ADHD (ayghri). Philosophy rests on 9 principles: YAGNI, plan before code, independent critique, diff audit, executable verification, idempotency, deterministic arithmetic.
Requires adaptation to your stack: CLAUDE.md instructions are tailored per project, skills go into .claude/skills/. No hard hardware or OS requirements — works wherever the used CLI agents run. No automated telemetry for hours — the 4500+ hour estimate is subjective.
// USE CASES
Best for: Developers using AI agents for production code
- Set up a controlled workflow for an AI agent in a production project
- Introduce independent plan critique and diff audit before commit
- Adapt curated skills (Superpowers, Ponytail) to your stack
New repositories every 30 minutes
REDDYX AI scans GitHub 24/7 and ships the best AI/ML/Web3 projects to Telegram.
Join on Telegram