Execution authority protocol
The protocol layer for governed AI execution.
Identity proves who the agent is. Access grants what it can reach. reht™ verifies whether a specific action is right to execute, then writes the receipt.
Built to sit between identity, policy, tools and execution.
The missing layer
Identity is not execution authority.
SSO, OAuth, OIDC, SAML, JWT and MCP Auth establish identity and access. They do not decide whether this action, in this state, with this evidence, should be allowed to change the world.
Who is acting?
User, agent, service account, organization and session context.
What can be reached?
Scopes, roles, credentials, grants and tool permissions.
Is this action right to execute?
Runtime check of authority, policy, evidence, state and risk.
Can the decision be proven?
Every execution handoff leaves an auditable decision record.
Protocol contract
One action envelope. One decision boundary.
01
Action
The proposed operation is made explicit before any tool, API or workflow is invoked.
02
Authority
The represented person, team or organization is bound to the action.
03
Policy
The relevant rules are evaluated at runtime, not assumed from login.
04
Evidence
Source validity, context and missing evidence are checked before action.
05
Decision
ALLOW, MODIFY, DEFER, DENY, STEP_UP or HALT.
06
Receipt
The outcome is written as an auditable execution record.
Mental model
Reasoning may be probabilistic. Execution authority must be governed.
Network play
Provider-neutral by design.
The same authority packet can govern a pull request, purchase requisition, CRM update, payment instruction, cloud deployment or MCP tool call. Adapters change. The boundary stays the same.
Positioning
Identity proves who the agent is. reht™ decides what the agent may do.
A light protocol layer for governed AI execution across agents, tools and enterprise systems.