Context engineering: give the model just what it needs

What you'll see in this lesson

1. Context engineering: what the model has in front of it

   Writing the question well isn't enough: the result depends on everything the model reads before answering.

2. Always loaded or only when needed

   The golden rule: what every task needs stays loaded; what's needed only sometimes is loaded on demand.

3. CLAUDE.md and AGENTS.md: the project manual

   They're text files in the project root that the agent reads at the start of every session.

4. Rules that load only where they're needed

   Split instructions by area of the project so the agent only reads the ones for the part it's working on.

5. Agent Skills: knowledge that loads in stages

   A skill is a folder with a SKILL.md and, if needed, scripts or templates the agent opens only when the task calls for them.

6. Memory: what the agent remembers between sessions

   Memory saves notes from previous sessions and loads them again at the start: it's useful, but you need to review it.

7. Documents and connectors: search instead of paste

   Instead of pasting everything, let the model search for what it needs in your files and apps when it needs it.

8. Measure before you trim: /context in Claude Code

   The /context command shows how much each part of the context takes up and suggests what you can optimize.

9. Summary: keep the fixed part minimal, the rest on demand

   Context engineering means giving the model exactly what it needs, at the moment it needs it.