Agent Manager

Who Uses It

• Operations leaders
• Technology leaders
• Approved agent users
• Administrators responsible for access and governance

Inputs

• Approved user records
• Approved agent records
• Skill and version metadata
• Permission rules
• Secure CDM API scopes

Processing

• Checks whether an agent, skill, or version is approved
• Applies permission boundaries before Secure CDM access
• Supports agent-aware workflow access
• Creates a foundation for monitoring and safety workflows

Human Controls

• Agent approval records
• Skill version checks
• Permission-based API access
• Administrative review of who can use which workflows

Security and Privacy Notes

• Designed to prevent unapproved agent access to private business data
• Does not require broad access when a narrower permission is enough
• Monitoring, alerts, anonymization, and redaction should be scoped around the workflow and deployment requirements

Outputs

• Approved or denied access decisions
• Permissioned API access
• Agent and skill governance records
• Roadmap monitoring and alert signals

Integrations

• Secure CDM
• Private Skills
• Approved local agent workspaces
• Approved managed agent environments

Implementation Pattern

• Start with lightweight approvals for specific users, agents, and skills.
• Limit access to the workflow APIs required for the task.
• Expand toward monitoring and safety controls as agent usage grows.

Results or Expected Outcomes

• Clearer permission boundaries for agent-assisted work
• Reduced risk from unapproved tools or stale workflow packages
• A stronger governance layer for private AI adoption

When to Use

• Approved users need agents or skills to work with private data
• The business needs governance around agent access
• Skill versions and permissions matter to operational safety

When Not to Use

• No agents or private skills are involved
• The task is one-time advisory work with no private workflow access
• The company cannot yet define approval rules

Related Systems