What is a CLI?
You've been typing prompts into a browser. A CLI is a different window to the same capability — and it unlocks things the browser can't do.
How context files work
Every time you start a new chat, the AI has no memory of you. A context file changes that — you write your briefing once, and the AI reads it every session.
Context hygiene
A CLAUDE.md that starts well tends to accumulate clutter. Context files degrade in predictable ways — here's how to recognize the signs and fix them before they start hurting your results.
Skills vs. commands
Skills and commands look similar but behave differently. One applies judgment. The other runs the same steps every time. Knowing which to use is the difference between a flexible tool and a reliable one.
Hooks as guardrails
A prompt asks the AI to do something. A hook makes the AI do it automatically, before or after every action. Hooks are how you enforce standards without relying on memory — yours or the model's.
Error logging
Fixing a mistake is not the same as learning from it. When you correct an AI error and move on, the lesson disappears with the session. Logging failures turns one-time corrections into lasting improvements.
How AI session memory works
AI tools don't remember conversations the way people do. Understanding how their memory actually works helps you structure sessions that stay focused and cost less.
Working with subagents
For complex tasks, Claude sometimes delegates work to subagents — separate sessions that each handle a focused piece. Errors in early stages compound through later ones. Here's what that means for journalism pipelines.