Everything up to here has been about one person, one project, one session. You open Claude Code or Cursor, you ask for something, you read the diff, you commit. That works. It stops scaling the moment the project outlives your memory of it, or the moment the same instruction has to be given for the fourth time.

This section is the layer above a single session. It covers the small set of command-line moves that make you faster than clicking, the files that carry your rules from session to session and how they stack up, the pointer that lets you come back after two weeks, the checks that run whether or not anyone remembers to run them, and what happens when one job is too big for one session. All of it exists so that the work you did on Monday is still usable on Thursday.

  1. The command line, properly: pipes, &&, exit status, and grep.
  2. Instruction files: global, project, and folder scope, and what a conflict actually does.
  3. Resuming across days: the pointer file that survives the gap.
  4. Automatic checks: hooks and CI, at a defined moment.
  5. Repeatable procedures: scripts, npm scripts, slash commands.
  6. Splitting work across sessions and agents: what survives a handoff.
  7. Connecting your tools to outside systems: what you grant when you plug something in.