Dive into agentic coding long enough and value of a separating the planning process from the implementation process becomes obvious. These planning artifacts have many names and varying levels of complexity (".plan", "SPEC.md", "PRD", etc.), occasionally even distributing skills to assist in creating and implementing them.
In this post I'll share a skill-free, lightweight version of this that has been working well for me lately.
This was surprisingly tricky to figure out since I had to consult a combination of QEMU docs, StackOverflow and GitHub issues, many of which had outdated arguments and options, or didn't apply for aarch64/arm architectures. Writing out the steps that worked for me as a guide in case they help others.
Recently, I've really gotten into tmux for managing all my terminal sessions/windows. Not so much for the panes, but more for keeping a highly contextual environment per project or task.
Most modern web stacks allow the "filtering" of requests via stackable/composable middleware, allowing you to cleanly separate cross-cutting concerns from your web application. This weekend I needed to hook into go's http.FileServer and was pleasantly surprised how easy it was to do.
This week, @bradfitz (of memcached fame) released groupcache at OSCON 2013. I'm already a big fan of memcached and camlistore, so I couldn't wait to download it and kick the tires.