consolesplayingconsoles connects physical hardware — retro consoles and other devices — onto a shared network, and exposes a dashboard (Pluto) and tooling that can reach those devices over SSH. Some of that tooling is driven by automation and AI agents. This document states the rules that protect the person whose hardware is on the network.

Connecting to a console or device is an explicit, per-command action — never a standing grant.

• Any connection to a physical device (e.g. ssh wii …) must be authorized for that specific command. Approving one task does not roll forward into permission to connect again later.
• Automation and AI agents must surface the exact command they intend to run on a device and obtain approval before each connection — including read-only ones.
• Capability is not consent. Passwordless key auth exists for convenience; it makes a connection possible, not permitted. The operator draws that line, per request.

• Device access defaults to read-only. Anything that writes to or changes a device is called out explicitly and confirmed first.
• Every command run against a device is shown to the operator — no silent or background access.
• Scope is the minimum needed for the task at hand and nothing more.

• One dedicated SSH key per host. Never share a single key file across devices — generating a key for one host must never overwrite another's. Per-host keys are wired via IdentityFile + IdentitiesOnly yes in ~/.ssh/config.
• Private keys are never committed. Neither are environment files with real data: only *.env.sample templates (zero real data) belong in the repo.
• Local agent memory, machine-specific config, and anything embedding an operator's home path stay out of version control (see .gitignore).

Found a way this network could expose or harm a connected device or its owner? Please open an issue describing the problem (omit any real credentials), or reach the project privately via the contact on @consolesplayingconsoles.