No agent-held secrets
Tokens are minted on demand and scoped to the agent’s manifest, connector, and approved operations.
Agents should not carry credentials or decide when they need approval. ARX connectors hold the credential, score the action, route manager review, and write the personnel record.
Every operation carries an owner, scope, risk class, approval rule, and audit outcome. Below the line, work proceeds. Above the line, the manager decides.
Tokens are minted on demand and scoped to the agent’s manifest, connector, and approved operations.
High-risk writes are paused before they reach the SaaS API. The approval is bound to a named manager.
The personnel record includes action, score, formula inputs, verdict, approver, and witness hash.
The same governance grammar applies whether the agent touches identity, ITSM, code, data, or security tools.
Low-risk reads proceed with record generation.
Scoped writes can require manager approval by operation.
High-risk actions can be blocked deterministically.
Termination invalidates connector grants.
Read workforce context, bind managers, and revoke access at termination.
Open incidents, draft changes, close duplicates, and pause destructive writes for approval.
Let agents triage signals while ARX supervises remediation scope and audit posture.
Govern draft artifacts, pull requests, reporting queries, and CRM updates by role and manager.
The agent asks the connector to perform a declared operation with runtime context.
agent → connectorARX evaluates operation risk, connector sensitivity, target sensitivity, and session behavior.
0-100 riskAllow, deny, or route to the named manager with diff and rationale.
policy decisionThe result lands in the personnel record whether the action was allowed or blocked.
hash-chainedWe will map the first cohort’s tools, writes, manager gates, and termination path.