Claudelab: a manually-gated alternative to Anthropic's agent 'Dreaming'

Claudelab is a personal workbench pattern designed to manage long-lived AI projects using Claude Code, tackling common problems like audit inflation, project pollution, and rule drift. It implements features such as explicit per-session limits on new memory files and a REJECTED archive for discarded ideas.