Skip to content
AI|Course

Customizing the agent and its environment · Advanced

Engineer Repository Context for Coding Agents

Coding agents are only as good as what they can see. Learn to write instruction files, make code easy to find, build fast feedback loops, package skills and subagents, and measure whether any of it helps.

7 lessons · 15 hr 15 min · Free

You will learn to

  • ▸Explain why context, not just the model, decides coding-agent quality
  • ▸Write concise instruction files that prevent repeated mistakes
  • ▸Choose between keyword, semantic, repo-map and structural code search
  • ▸Build fast, readable feedback loops and connect knowledge with MCP
  • ▸Package procedures as skills, isolate exploration in subagents, and benchmark your results

Before you start

  • ·You use a coding agent (Claude Code, Cursor, Copilot, Aider or similar) on real code
  • ·Comfortable with Git, a test runner and the command line
  • ·Helpful: the Production Agents course, for the trust and security lessons

Syllabus

  1. 01Why repository context mattersWhat the agent sees, and why more context is not better.45 min
  2. 02Instruction files: AGENTS.md and friendsWhat to write, what to leave out, and how to keep it alive.1 hr 30 min
  3. 03Code search and retrievalGrep, embeddings, repo maps and structural search compared.2 hr 30 min
  4. 04Tools, feedback loops and MCPFast verification, readable failures and external knowledge.2 hr
  5. 05Skills and subagentsProgressive disclosure and context isolation.1 hr 30 min
  6. 06Evaluating and maintaining contextPrivate benchmarks, controlled comparisons and context rot.2 hr
  7. 07Capstone projectEngineer a real repo for agents and prove the improvement.5 hr