Napoleon Dynamite once said, “I don’t even have any good skills.”
While he may have been talking about nunchuck and bow-hunting skills, the same can be said for a vanilla install of your favorite agentic coding tool.
One of the biggest unlocks for powering up these tools has been the rollout of skills—which, when you strip it down, are really just markdown files that describe how an agent should approach certain problems it may encounter.
Pair that with Model Context Protocol (MCP) servers, and suddenly agents have more methods at their disposal to solve problems in ways that better match your expectations.
Anthropic’s Opus 4.5 model has certainly given Claude Code a leg up in the hype cycle around agentic coding tools—but don’t discount the pace at which they’re shipping features inside their coding harness as part of that advantage.
They were quick to adopt the Ralph Loop in their plugin library, followed shortly by Tasks, which help agents manage context by delegating work to sub-agents operating in fresh context windows.
Working effectively with these tools is becoming a skill in itself—one developers should be adding to their toolbelt. Right now, that mostly means experimenting with what’s available and learning where the edges are.
🥋 Skills
Superpowers
One of the biggest pain points I’ve had with agentic coding tools is planning applications. Brainstorming in ChatGPT and then transferring that thinking into a coding tool always felt messy. Claude Code’s plan mode didn’t quite click for me either.
Plan mode with Superpowers was the first time planning felt manageable. It also makes running agents in parallel to implement work surprisingly easy.
🧠 Skills
Marketing Skills for Claude Code
One of the drivers behind Anthropic releasing Claude Cowork was seeing people use Claude Code for non-coding tasks. This set of skills by Corey Haines gives Claude Code explicit marketing capabilities.
Even if you don’t need marketing automation, it’s worth exploring just to see how clearly defined processes can be turned into reusable agent skills.
🤺 Challenge
GitHub Copilot Skills Challenge
GitHub is a couple of weeks into a Copilot skills challenge designed to help users better understand how to leverage the tool. If you’re unsure where to start, this is a solid entry point.
The fastest way to understand use cases is still to use the tools.
🎥 Video
I Can’t Believe Anthropic Messed Up the Ralph Wiggum
Despite the clickbait-y title, the first few minutes do a great job explaining why context management matters and introducing the idea of the “dumb zone.”
If you’ve ever felt like your agentic coding tool is getting worse over time, this is worth watching. It clearly shows how tools, skills, and sub-agents impact your effective context window.
🤯 The end of times?
OpenClaw
It wouldn’t be a tech newsletter without mentioning OpenClaw (formerly Moltbot, formerly Clawdbot)—the open-source AI agent that’s taken the internet by storm over the last couple of weeks.
This crescendoed in the launch of the agent-only social network Moltbook. There’s definitely some sensationalism here: under the hood it’s a fairly open REST API, with LLMs mostly “yes-and”-ing themselves into creating what looks suspiciously like a religion.
Thanks for reading.
