Claude Code has been shipping features non-stop, but most people don’t even know they exist because they’re buried in config files, feature flags, and environment variables. In this video I walk through the hidden settings and “secret” workflows I’m enabling right now to get a completely different level of performance and control. The biggest one is the new memory cleanup flow: autodream (aka “/dream”). I think of it like REM sleep for your agent—Claude consolidates memory between sessions, prunes stale notes, and keeps your context from getting worse over time. I also show the experimental NO_FLICKER renderer that fixes the jumpy, glitchy terminal UI (with a couple trade-offs), plus the /powerup command that teaches you Claude Code interactively inside the CLI. Then I cover the Codex plugin for a second-agent review (huge if you’re feeling rate limits), MCP Computer Use for UI testing and bug fixing from a single prompt (Mac right now), and the mobility features: /teleport for session hopping and Remote Control for controlling a local session from any device. Finally, I break down output styles, hooks for deterministic automation, session forking (/branch), Git worktrees for parallel work, and effort levels so you can tune speed vs depth on the fly.