SilverBullet spaces are scriptable environments. This is where a lot of SilverBullet’s power and flexibility comes from. This page lays out the trust model that follows from that, so you can configure a deployment that matches who you’re sharing it with. For a more basic walk-through of concrete setups, see Security Profiles.
You need to trust people with write access. Anyone who can write to a space can author Reference/Space Lua and other dynamic content that runs in the browser of anyone who later opens that space.
Each space resolves to one of three access levels for a given visitor, plus a separate admin flag:
none: Not visible. Requests are refused.read: May view content, will run all scripts and queries locally. Cannot modify anything or reach any capability endpoint.write: Full access to the space’s content and capabilities.Admins get write on every space, plus the /.dashboard management UI.
write really meanswrite is not just “can edit files” in the narrow sense — a write member can create content that runs, unattended, as whoever else opens that space. That includes admins.
The practical rule: grant write only to people you’d trust to act as any member of every space they can reach. If someone doesn’t clear that bar, don’t add them as a write member of a space other trusted people also use, give them their own space, on its own (sub)domain instead (see Security Profiles#Software team). SilverBullet does not offer an in-between “can edit, but can’t script” role within a single space.
A few endpoints can reach beyond the space’s own content, and all require write:
/.shell) — runs any command on the server. (Off by default)/.proxy, exposed to Lua as net.proxyFetch) — fetches an external URL through the server, on the server’s network./.runtime) — evaluates Lua and scripts via a headless browser instance, see Integrations/Runtime API.