Here are my tips for getting the most out of Claude Code, including a custom status line script, cutting the system prompt in half, using Gemini CLI as Claude Code’s minion, and Claude Code running itself in a container.
Nice list. For tip 27, the killer feedback loop I've recently discovered for building Web apps is to use the Chrome DevTools MCP server. This gives Claude the power to visually inspect the app in a running dev server as well find and fix errors that only happen under real condition on the client side.
Your "AI context is like milk" metaphor really lands. The handoff document approach is brillaint—instead of letting Claude auto-compact into a black box, you're explicitly building institutional memory that the next agent can parse. What's intresting is this mirrors how senior engineers do knowledge transfers in oncall rotations, except now we're doing it mid-session between AI instances. The tmux-based testing pattern also generalizes nicely beyond git bisect.
Awesome tips! I’m a very casual user, but intend to dig in deeper. This list gives me lots of things to try out. Thanks for sharing!
Nice list. For tip 27, the killer feedback loop I've recently discovered for building Web apps is to use the Chrome DevTools MCP server. This gives Claude the power to visually inspect the app in a running dev server as well find and fix errors that only happen under real condition on the client side.
Your "AI context is like milk" metaphor really lands. The handoff document approach is brillaint—instead of letting Claude auto-compact into a black box, you're explicitly building institutional memory that the next agent can parse. What's intresting is this mirrors how senior engineers do knowledge transfers in oncall rotations, except now we're doing it mid-session between AI instances. The tmux-based testing pattern also generalizes nicely beyond git bisect.