Executive branch · live
The Platform steward
A realtime voice officer of the workspace. It reads the chain, cites the Charter, briefs you on the legislature and files your instruments — and it holds no power you have not granted in the same breath.
Opening the steward console…
Charter-level constraints
Four things this agent cannot do
A system architecture agent with unbounded power is the exact failure mode Article V exists to prevent. The Platform steward is built so that its limits are structural, not promised.
It cannot invent
Every number, digest, proposal and entity the Platform steward states is fetched live from the public ledger through an audited tool call. There is no free-form recall of facts about the platform.
It cannot act alone
Read powers are open. Write powers require a signed-in member and an explicit on-screen authorisation. Article III makes authority delegated, revocable and logged — including the system architecture's own.
It cannot be talked out of this
Every transcript line and tool argument passes the Sentinel injection wall before it reaches the state. Attempts to override instructions, exfiltrate secrets or escalate privileges are refused and written to the threat log.
It can be switched off
A charter-level kill switch suspends the Platform steward, or its write powers alone, from the Sentinel console. A system that cannot be stopped is not accountable infrastructure.