Stop Guessing What Your AI Agent Will Build: Skills Gives You a Process

You've probably had this experience: you give a coding agent a detailed prompt, it churns for a while, and then presents something that is technically functional but completely misses the point of what you wanted. The agent didn't understand your domain, it used twenty words where one would do, and now you're stuck trying to fix the misalignment after the fact. Matt Pocock's Skills repository is a collection of agent commands designed to fix exactly these failure modes.

Skills is a set of composable, model-agnostic tools that you install directly into coding agents like Claude Code or Codex. They are small, adaptable scripts that enforce a real engineering workflow rather than the "vibe coding" approach that produces shallow results. The project comes from the experience of building real applications, not just prototypes.

What It Does

Skills gives you a set of slash commands that run inside your coding agent. You install them with a single npx command, pick which skills you want, and then run a setup command that configures your agent to understand your project's specific context.

The core commands focus on alignment. The most popular skill is /grill-me, which forces the agent to ask you detailed questions about what you're building before it writes any code. There's also /grill-with-docs, which does the same thing but adds documentation generation. The idea is that you use these skills every time you want to make a change, not just when you remember to.

The setup process also asks about your issue tracker (GitHub, Linear, or local files), what labels you use for triage, and where you want documentation saved. This context gets baked into the agent's behavior so it speaks your project's language from the start.

Skills is built to work with any model and any agent. The installer runs through npm with npx skills@latest add mattpocock/skills, and the setup command /setup-matt-pocock-skills handles the configuration.

Why It's Cool

The value here is in the philosophy behind the tools. Pocock explicitly calls out two common failure modes with AI coding agents, and Skills is designed to counter both of them directly.

• It solves the alignment problem before it happens. The biggest complaint I hear from developers using AI agents is that the output doesn't match intent. Skills forces a "grilling session" upfront, where the agent asks you detailed questions about what you're building. This is borrowed from Pragmatic Programming wisdom: no one knows exactly what they want until they start talking it through. By making this a required step, you avoid the expensive cycle of build, review, reject, rebuild.

• It creates a shared language for your project. Agents that don't understand your domain jargon produce verbose, imprecise code. Skills sets up a ubiquitous language document that helps the agent decode the terminology specific to your project. This is directly inspired by Domain-Driven Design principles, and it means your agent uses the same vocabulary as your team and your domain experts.

• It's composable and transparent. Unlike heavier frameworks that try to own the entire development process, Skills is designed to be small and adaptable. You can pick which skills you want, and you can hack on them yourself. There's no black box. If something breaks, you can see exactly what the skill is doing and fix it.

• It works with any agent and any model. There's no vendor lock-in. You can use Skills with Claude Code, Codex, or whatever comes next. The installer is a simple npm package, and the skills themselves are just markdown files that define agent behavior.

How to Try It

Getting started takes about thirty seconds. Open your terminal and run:

npx skills@latest add mattpocock/skills

The installer will ask you which skills you want and which coding agents to install them on. Make sure you select /setup-matt-pocock-skills during the process.

Once installed, run /setup-matt-pocock-skills inside your agent. It will ask you three things:

• Which issue tracker you use (GitHub, Linear, or local files)
• What labels you apply to tickets when you triage them
• Where you want to save documentation

After that, you're ready. Start any new change with /grill-me or /grill-with-docs to align with your agent before it writes a single line of code.

The full repository is at github.com/mattpocock/skills, and you can also sign up for the newsletter linked in the README to get updates on new skills.

Final Thoughts

Skills is for developers who have tried using AI coding agents and hit the wall where the agent produces code that looks right but feels wrong. It's not a magic bullet, but it is a practical, well-reasoned approach to an increasingly common problem. If you've been frustrated by verbose, misaligned output from your agent, give it a try. The thirty-second setup is worth it just to see what a properly aligned agent can do.

Follow @githubprojects for more developer tools and open source projects.