5. Context Engineering

5.1 What is context engineering?

Prompt engineering focuses on the instruction.

Context engineering focuses on the broader information and capabilities available to an AI system.

Prompt
+
Project instructions
+
Relevant files
+
Skills
+
Tools
+
Retrieved information
+
Previous decisions

5.2 Why it matters

Suppose an agent is asked to modify an API.

Useful context:

API implementation
API tests
Database model
Authentication rules
Project instructions

Potentially unnecessary context:

Old build logs
Unrelated frontend components
Generated files
Large dependency directories

5.3 Persistent context

Project instructions can encode stable rules:

AGENTS.md

Examples:

How to run tests
Architecture rules
Coding conventions
Security requirements
Git rules

5.4 On-demand context

Skills are useful when knowledge is specialized.

General project rules
       +
Relevant skill
       +
Relevant files

This is preferable to putting every procedure into one giant instruction file.

5.5 Context hierarchy

A practical hierarchy:

Global guidance
      |
      v
Project guidance
      |
      v
Directory-specific guidance
      |
      v
Relevant skill
      |
      v
Task-specific context

OpenCode supports AGENTS.md discovery at global and project/directory scopes. See the official instructions documentation for the exact discovery behavior.

Exercise

Design the context for an API testing task.

List:

Persistent rules
Relevant files
Relevant skill
Required tools
Information that should not be included