muse
Installation
SKILL.md
Muse toolkit — one skill, four lanes
Everything here is specific to running on a Muse VM. The platform breaks normal Linux assumptions (the box is a container that gets replaced, DNS lies, egress is gated per category), and each lane below is a verified playbook for one job. Facts are dated 2026-11-02; this machine's own records beat this file when they disagree — re-run the probes instead of trusting stale text.
| You want to… | Read |
|---|---|
| know what survives a redeploy, where to store things, how env/users/ssh config really resolve | references/platform.md |
| reach the internet or another machine, understand a blocked/reset connection, pick a permission to ask for | references/egress.md |
| join a tailnet, accept or originate SSH, build remote access into this box | references/join.md |
| follow a complete, worked join (fresh box -> verified fleet member) step by step | references/example-fresh-box.md |
| keep a daemon alive, schedule work, or react to a condition — systemd vs cron vs hooks | references/automation.md |
| stand up a FRESH Muse VM as a join/e2e target (the muse-test pattern) | scripts/setup.sh (self-contained installer; secrets via a 600 .env it templates for you). Call it by full path — the skill folder names the target; there is no dispatcher. Phased like the replit setup: --sshd, --tailscale, -j/--join, --verify, --user <name> (stack service user; env RS_SSH_USER, flag wins, default root — setup prints a warning for root; --user runner is the fleet-proven isolated mode), and an interactive arrow-key menu (bare run on a TTY, or --menu); a bare run without a TTY does the full setup |