Articles
Original writing on getting more out of AI tools.
What is an MCP server? A practical guide
An MCP server exposes tools, resources, or prompts through a standard protocol so an AI application can discover and use external capabilities.
How to vet AI agent skills before installing them
Treat an AI agent skill as both an instruction package and a software dependency: inspect what it says, what it runs, what it can access, and how it updates.
How to add an MCP server to Claude Code safely
Add remote HTTP or local stdio MCP servers to Claude Code, choose the right scope, protect credentials, verify the connection, and test with least privilege.
Claude skills vs. connectors vs. plugins: what is the difference?
Skills teach Claude a repeatable method, connectors provide governed access to apps and live data, and plugins package related capabilities for installation and sharing.
AI agent skills vs. MCP servers: which should you build?
Use an agent skill to package reusable know-how and workflow instructions. Use an MCP server when an agent needs live, governed access to external data or actions.
Claude Code skills vs. slash commands: what changed
Custom commands and skills can both create a slash-invoked workflow in Claude Code. The important choice is how the workflow is discovered, shared, and permissioned.
10 best Claude skills for developers in 2026
A useful Claude skill solves one recurring engineering job, is easy to inspect, and saves more time than it creates in setup and review.
How to install Claude skills in Claude and Claude Code
Claude skills can live in your Claude account, your local Claude Code setup, or a repository. Install them where the sessions that need them can load them.
How to create an AI agent skill: a practical SKILL.md guide
Build a portable AI agent skill from one repeatable job: a precise description, concise instructions, focused resources, and tests that prove it works.
What are AI agent skills? How SKILL.md works
AI agent skills package instructions, scripts, references, and templates into portable folders an agent loads only when the task calls for them.
Point your agent at the catalogue: the LLM Mart MCP server and API
The whole public catalogue is an MCP server and a REST API, so your agent can search skills, tools and slash-commands as native tools. Setup is one config block. Plus a private vault that carries your own prompts between machines.
AI content SEO: how to publish pages that earn their place
AI made publishing cheap, which is exactly the problem. What separates a page worth ranking from a competent summary of the first ten results.
Prompt testing: build a lightweight eval set
A prompt that works once isn't a quality system. Five cases, an observable rubric, and a regression set will tell you whether a change helped.
The version bump that took down an app (and how it self-healed)
One character of YAML, four pods that never started, and two safety nets I didn't know were holding. Every restart is an audit. Schedule them before they schedule you.
What counts as proof
"Verify your work" isn't an instruction. It's a mood. Here's the version that's an instruction. Verify with a different mechanism than the one that made the claim.
Prompt Testing: Build a Lightweight Eval Set
A prompt that works once may still fail in production. A lightweight eval set gives you repeatable cases, a clear rubric, and a way to see whether a prompt change actually improved the workflow.
How to Choose an AI Tool: A Practical Evaluation Checklist
The best AI tool is not the one with the longest feature list. It is the one that solves a defined job reliably, fits the workflow, handles data appropriately, and remains useful after the novelty wears off.
How to Research with AI Without Losing the Source Trail
Use AI to speed research without losing trust. Learn to find primary sources, verify claims, preserve uncertainty, and keep an auditable source trail.
How to Write Better AI Prompts: A Practical Framework
Better prompts aren't magic wording. They're short briefs that hand the model a task, the context it can't infer, the limits, and a quality bar.
The deploy that merged but never deployed
A green PR, a controller reporting success, and not one line of the new code running
Ship a prompt like code: a five-case eval harness you can build in an hour
You wouldn't ship a function you ran once. Here's the smallest evaluation setup that catches real regressions — five cases, three graders, and a rule for when to add a sixth.
The instruction file is the highest-leverage file in your repo
AGENTS.md, CLAUDE.md, copilot-instructions.md and .cursor/rules all solve the same problem. Most of them are written badly. Here's what changes agent behaviour and what's decoration.
Prompt injection is not an XSS problem
Sanitizing output protects your page. It does nothing for an agent that reads a poisoned README and then runs a command. A practical model of the threat, and what actually helps.
Prompt, skill, workflow, recipe, template, agent: a working taxonomy
The words are used interchangeably and they shouldn't be. Six categories, what actually distinguishes them, and how to tell which one you're holding.