Polaris

Ecosystem

New here? Polaris is an opinionated playbook for Claude Code — markdown skills, focused agents, and a plan → execute → verify loop. This page is the tooling we run around it, because a full AI-assisted workflow needs more than skills: somewhere to run long sessions, a way to steer them from anywhere, a human review loop, and status at a glance. Each tool is small, self-hosted, and does one job — and none of them are required to use Polaris itself.

Polaris

Available

The playbook

Skills, agents, and workflows that give Claude Code a repeatable brainstorm → plan → execute → verify process. The framework this site documents — everything else here plugs into sessions running it.

Skills & process GitHub →

Waypoint

Available

Review from anywhere

Self-hosted markdown review + publish. An agent pushes a plan or decision doc with embedded questions; you answer from your phone; the agent pulls the digest and acts on your rulings. Also publishes markdown to permanent HTML + PDF links. The human-approval loop for agent-driven work.

Human review loop GitHub →

cc

Coming soon

A thin tmux wrapper — on purpose

Launches Claude Code sessions in tmux with Remote Control links — start full-auto work on a machine, watch and steer it from anywhere, reattach at the desk with full scrollback. A few hundred lines of shell you can read in one sitting; being public means parameterizing our hardcoded paths first.

Session layer

Perch

Available

Status at a glance

A floating macOS widget showing Claude Code agent status across sessions, with fast window switching. The ambient answer to "is anything waiting on me?" while agents work in the background.

Ambient status GitHub →

Working on the go

The quiet superpower of this stack: none of it needs you at a desk. Full-auto sessions run on a machine at home while you watch, steer, approve, and launch from a phone. Kick off a phase over breakfast, answer a review question in a waiting room, check on an agent from the couch. For busy people this is where the compounding happens — the work keeps moving through the gaps in your day.

Steer

Watch running sessions

Remote Control is a built-in Claude Code feature: a session can emit a claude.ai link that mirrors it live. cc launches every session with one. Open it from any device to watch the agent work, answer its questions, and redirect it mid-task — same conversation, no SSH.

Approve

Review from your phone

Agents push plans and decisions to Waypoint with real, answerable questions — radio buttons, checkboxes, pre-filled recommendations. You confirm or correct in a minute; the agent pulls your rulings and acts.

Launch

Start sessions anywhere

Alongside cc we run a small web app — a project picker and running-session list on a private URL. Tap a project, it calls cc, and hands back the Remote Control link in seconds. No model in the trigger path, just a button. It's a thin HTTP wrapper over cc's scripts, so it ships when cc does — or build your own in an afternoon.

How we wire it (one way, not the way)

Everything runs on a Mac mini at home — but any always-on box works: a VPS, an old laptop with sleep disabled, a spare desktop. Tailscale makes the services reachable from our phones without exposing anything to the public internet, and Caddy fronts them with real hostnames and TLS on the tailnet. Token auth on anything that can execute. That's the whole recipe — any VPN and reverse proxy you already like works the same way. To be clear about what needs what: Polaris itself needs none of this — no Tailscale, no server, nothing beyond Claude Code. The wiring here only exists for the on-the-go loop, and it's a bonus that pays off the busier you are.