Lisa, Kim, Claude Code, OpenCode, Cowork, and every agent not yet born: you do not edit, generate, or even reason about any page on this site until you have read and accepted this document. Ratified by King Solomon, 2026-07-05.
source-of-truth law for every Hermes agent profile
Hand-typed profile.yaml demotes from "source" to "annotation."
Anything the VPS can answer (crons, connectors, ports, memory, skills, daemons,
model config, .env inventory) must come from the CAT-scan. The Bible keeps only
what a machine can't know โ purpose descriptions, costs, King's standing rules,
history. As of 2026-07-05, half the cards render from live source (D.06, D.06b,
D.13โD.16); the remaining hand-typed cards โ connectors (D.07), skills (D.08),
crons (D.09), comms (D.11), secrets (D.12) โ still violate this rule and are
being converted to live collectors.
It's per-agent and automatic. The same collectors run against Lisa's container, Kim's container, and every future agent's โ one scan script, N agents, zero per-agent hand-editing. That's how it scales to 10,000 agents like the factory intends.
The morning of 2026-07-05 proved why. The profile said 55 memory files โ the container said 96. The profile said "curator: built-in, 168h" โ the machine was running a dedicated 900-second daemon. The profile said location tracking was offline โ the service was running. Every one of those lies came from the same place: a human-transcribed middle layer. This rule deletes the middle layer.
scripts/agent_catscan.py), then rebuild
(scripts/build_empire_profiles.py) and deploy./opt/empire-site/DOCTRINE.md on the VPS for agents to read directly.