AI Coding on Top AI Skills

DeepEval: Evaluate LLM Outputs with Your AI Coding Agent

Mon, 13 Jul 2026 00:00:00 +0000

If you build AI-powered features, you’ve felt the pain. You ship a prompt change, and suddenly the model starts hallucinating in production. Or you switch from GPT-4 to Claude, and the output format breaks silently. DeepEval solves this — it’s the testing framework for LLM outputs, now available as an AI agent skill.

What It Does

DeepEval treats LLM evaluation like unit testing. Instead of manually eyeballing outputs, you write test cases and metrics — hallucination score, answer relevancy, faithfulness, toxicity, bias, and more. Your agent runs these against your model’s responses and tells you what broke and why.

Best AI Coding Assistants in 2026: A No-Nonsense Roundup

Tue, 16 Jun 2026 00:00:00 +0000

The Problem + Selection Criteria

You open Twitter, LinkedIn, or any developer forum in 2026 and there are seven different AI coding tools claiming to be the best. Claude Code runs in your terminal. Cursor is a full IDE. Copilot lives inside VS Code. Codex CLI is open source. Windsurf has its own editor. Continue plugs into anything. Amazon Q wants your AWS bill.

They all generate code. They all claim to save time. But they make different trade-offs — in context handling, upfront cost, workflow integration, and the kind of projects they handle well. Picking the wrong one means fighting your tool instead of your code.

Claude Code vs Cursor: Which AI Coding Tool Should You Use?

Mon, 15 Jun 2026 00:00:00 +0000

The Honest Verdict Up Front

Claude Code wins if you live in the terminal. Cursor wins if you want AI baked into a full IDE. Neither is objectively better — they serve different working styles. After spending months with both, I reach for Claude Code when I want speed and script-like interactions, and Cursor when I need visual context and multi-file editing with less mental overhead.

The real answer? Most serious AI-coded projects benefit from having both available.

I Let AI Write My PRD for a Week Here's What Happened

Fri, 12 Jun 2026 00:00:00 +0000

Why I Went Down This Rabbit Hole

It was a Tuesday afternoon, and I was staring at my third blank PRD template of the month. Same cursor blinking in the same “Problem Statement” field. Same feeling that I should be writing code instead of documenting why the code matters.

I work on a small team. We ship fast, which means documentation is always the first thing to slip. PRDs end up as Notion scraps or Slack threads titled “that thing we discussed.” When I found to-prd on skills.sh — described as a tool that synthesizes conversation context into a structured PRD and publishes it to your issue tracker — I installed it within 30 seconds. The promise was simple: stop writing requirements, start shipping. I’d already tried Grill Me, another Matt Pocock skill, for plan review. to-prd felt like the natural companion.