12 · Agents

The agent fleet that extends Twin capabilities.

Agents are specialized AI participants that operate within governance. They are not autonomous actors — they execute under the authority of the Twin and the Governance layer.

---

Agent Taxonomy

                    AGENTS
                       │
         ┌─────────────┼─────────────┐
         │             │             │
    DOMAIN AGENTS  SPECIALIST    SYSTEM AGENTS
         │         AGENTS            │
         │             │             │
    Sales Agent   Research Agent  Orchestration
    Finance Agent Analysis Agent  Scheduling
    Support Agent Drafting Agent  Monitoring
    Account Agent Integration     Reconciliation
    Operations    Agent           Memory

---

Domain Agents

Agent	Domain	Capabilities
Sales Agent	Sales	Pipeline analysis, deal scoring, next-best-action
Finance Agent	Finance	Revenue analysis, billing review, collection priority
Support Agent	Support	Ticket triage, SLA prediction, escalation routing
Account Agent	Account Success	Health analysis, renewal risk, expansion opportunity
Operations Agent	Operations	Workflow optimization, resource allocation

Domain Agent Rules

• Scoped to their business domain
• Reason and recommend within their domain
• Escalate cross-domain work to the orchestrator
• Feed learnings back to the Spine

---

Specialist Agents

Agent	Specialty	When Used
Research Agent	Deep investigation	When context is insufficient
Analysis Agent	Quantitative analysis	When numbers need interpretation
Drafting Agent	Content creation	When output needs to be produced
Integration Agent	System integration	When external systems need interaction

Specialist Agent Rules

• Invoked on demand
• Receive focused context for their task
• Return results to the calling agent
• Do not maintain persistent state

---

System Agents

Agent	Function
Orchestration Agent	Coordinates multi-step workflows
Scheduling Agent	Manages time-based triggers
Monitoring Agent	Observes system health
Reconciliation Agent	Verifies execution outcomes
Memory Agent	Manages organizational memory

System Agent Rules

• Operate at the platform level
• Not directly visible to users
• Enforce governance and policies
• Fully auditable

---

Agent Communication (ACP)

Agents communicate via the Agent Communication Protocol:

interface ACPMessage {
  from: AgentRef;
  to: AgentRef;
  workItem: WorkItemRef;
  context: ContextSnapshot;
  intent: string;
  constraints: Constraint[];
  deadline?: string;
}

ACP Rules

• Hand-offs are logged in the Spine
• Context is transferred, not lost
• Receiving agent can accept, reject, or delegate
• High-risk transfers require governance approval

---

Agent Security Boundary

Canonical Spine
      ↓
Projection (role + authorization)
      ↓
Context Assembly (relevance + token budget)
      ↓
Agent Context (sandbox)
      ↓
Agent Execution

Agents operate against authorized projections, never unrestricted canonical Spine.

---

Agents Connect to

Layer	Relationship
Twin Workbench	Twin delegates to agents
Work Allocation	Work assigned to agents
Governance	Agent actions gated by governance
Communication Bridge	Agent communication via ACP
Execution	Agents execute via Execution Fabric