Add unitreex: natural-language control of a real Unitree humanoid

[richard7463]

What this skill does

unitreex lets a Bankr agent drive a real Unitree G1 humanoid robot with plain language — "wave hello", "shake hands", "sit down" — via a small bridge server the user runs on the robot's LAN.

X timeline (bankr) ──HTTPS + Bearer──▶ tunnel ──▶ bridge (FastAPI, whitelist + FSM gating) ──DDS/LAN──▶ Unitree G1

Why it's a good fit

• Embodied agent primitive: agents already move money and post; this gives them a body. On-timeline intent becomes real-world motion.
• Safety at the edge: the agent can ask, but the bridge decides — hard action whitelist, FSM/battery gating, fail-closed auth, confirm-required dangerous actions. The SKILL.md encodes a strict safety contract (never invent actions, never bypass a 422).
• Zero-hardware onboarding: the bridge ships mock + MuJoCo sim backends, so users can wire the full path and record demos before the robot arrives.

Files

• unitreex/SKILL.md — agent instruction layer (setup, endpoints, action whitelist, safety contract)
• unitreex/catalog.json — schemaVersion 1 catalog entry with demo + setup steps
• unitreex/unitreex.png — 512px logo

Testing

Bridge repo: https://github.com/richard7463/unitree-skill — includes an end-to-end smoke test (wave succeeds, dangerous action returns 422, bad token returns 401). Live site: https://unitreex.club/