This is a partial user and operations guide. The full guide is available to platform subscribers and active users only. We publish this public selection so evaluators get an initial sense of capabilities and workflows. Private guide content may change with ongoing product improvements.
Governance
The architect proposes. The human approves. The gated arm applies. The service never provisions on its own.
Approving in the queue is applying — there is no second step. Destructive actions require scope in telemetry (require_scope), not a hidden button.
SOCIETY_GOVERNANCE_ENFORCE is OFF by default. Turning it on is an environment decision. The Console toggle in Dev is in-memory: it does not survive a container restart.
Kill-switches (default off): SOCIETY_SPAWN_ENABLED, SOCIETY_RAG_WRITE_ENABLED, SOCIETY_SKILL_PUBLISH_ENABLED, SOCIETY_EVALUATOR_AUTO_APPROVALS_ENABLED.
The evaluator also has a master switch, SOCIETY_EVALUATOR_ENABLED, on by default: when off it runs none of its automatic cycles and never calls the LLM judge. It is flipped live from the Console (Admin → Dev → «Qué está encendido en la sociedad»), like the other switches.
Roles (claim roles from the OIDC ID token → scopes):
| Role | Can, in short |
|---|---|
admin | Operate the society, SOUL/prompts, users; not dev:write |
operator | Approve, deploy, governance; no delete or prompt reconfig |
member | Conversation UI; :own reads |
viewer | Society read, no approval queue |
service | M2M (gated arm, graph emitters) |
dev:write and * are superadmin-only: no matrix role grants them.