Human OS

Claude coach Interrogate this architecture Open a Claude conversation primed to examine the sources of truth, privacy boundary, human-agent handoffs, and pace of the collaboration. Discuss it with Claude

What it is

Human OS is an early-stage, two-week launch cockpit for a creative collaborator. It is designed to let useful work begin without asking the collaborator to become a project manager or learn an agent stack first.

The shared surface is deliberately small: Linear holds current work, decisions, owners, and short updates; a private Git repository holds research, source records, and evidence; Excalidraw is available when a picture explains the system better.

What I built

  • A dedicated Linear workspace so the collaboration has its own privacy boundary.
  • A five-phase project, initial decision backlog, labels, views, issue template, and onboarding guide.
  • A simple default workflow, with labels describing the kind of attention needed rather than adding bespoke process states.
  • Idempotent setup through the Linear API, followed by a live read-back of the resulting workspace.
  • Explicit boundaries for human decisions, agent-ready work, private material, and participant data.

The project remains deliberately provisional. Discovery and human priorities determine which two or three interventions are worth testing; the machinery does not set the pace.

Read the build note →