This is a workshop to learn how to use some advanced context engineering techniques with Claude Code.
Watch the video: https://hlyr.dev/ace (or read the post)
More links and info: https://github.com/ai-that-works/ai-that-works/tree/main/2025-08-05-advanced-context-engineering-for-coding-agents
Want to discuss these techniques? Join us in https://humanlayer.dev/discord
- Claude Code acount and logged in on your workstation
- A way to view/edit markdown files (Cursor, VSCode, vim, go-grip, etc)
- A chosen issue to work on, ideally on an OSS repo that you have some familiarity with / understanding of
COST NOTE - we recommend using Opus for this workshop as its the best at reliably understanding and working in large codebases.
- familiarity with managing claude code subagents and commands
You should pick an issue to work on. It should be somewhat small just ot help you focus on learning the system.
You can use these techniques in your own private repos, but if you want help/etc I recommend using an open source repo that you have some familiarity with / understanding of.
A good bar to set, is to pick the OSS repo for a tool you use often enough that if you read a bug ticket, you can understand and reproduce the issue yourself.
WHatever issue you choose to work on , you will need to clone the repository to your workstation. This will be your "working repository"
Take the issue you want to solve and get it into a .txt or .md file in your working repository. You can copy the text from github, you can use the gh cli, you can prompt claude to use an MCP, whatever you want.
follw the instructions at https://humanlayer.dev/docs to install codelayer
open a new codelayer session, set the working directory to your repo, and type /cl:research_codebase
/cl:research_codebase - we are working on the issue in the issue.txt file, please read the issue and research the codebase to understand how the system works and what files+line numbers are relevant to the issue. Do not make an implementation plan or explain how to fix.
when you submit, you'll be prompted to bootstrap the core prompts and agents into your current repo
/create_plan - we are working on the issue in the issue.txt file, we've done the following research: PATH_TO_RESEARCH_OUTPUT.md
create a plan to fix the issue. YOUR ADDITIONAL INSTRUCTIONS HERE
/implement_plan - we are working on the issue in the issue.txt file, we've done the following research: PATH_TO_RESEARCH_OUTPUT.md
we've created a plan to fix the issue. PATH_TO_PLAN.md
please implement the plan. YOUR ADDITIONAL INSTRUCTIONS HERE
From here its up to you - you can borrow the /commit command from github.com/humanlayer/humanlayer/tree/main/.claude/commands/ to generate a commit message, you can push to a fork, and you can use /describe_pr to generate a pr description. (You might need to tweak the describe pr or go get a template if not present yet - ignore any error about "thoughts setup incomplete")