Unclassified
09 · Spine Agent Execution
A Spine Agent is a governed organizational actor — identity, role, authority, AI continuity, business context, OODA-derived plan, capability, governance, execution, verification, continuation.
A Governed Organizational Actor
A Spine Agent is not a chat persona. Not a tool wrapper. It is an organizational actor with identity, Work state, evidence, and continuity that operates over Business Context inside the AI Workbench.
Identity
↓
Role
↓
Authority
↓
AI Continuity
↓
Business Context
↓
Observe
↓
Orient
↓
Plan
↓
Capability (Fabric: what can be done?)
↓
Governance (may it be done? → approval → authorization)
↓
Execute (Hermes: execute what has been authorized)
↓
Observe Outcome
↓
Verify
↓
Evidence → Shared Work → Continue
Every agent action is authorized, traced, and reconcilable back to the Spine. No free-floating tool calls.
What Defines a Spine Agent
| Property | Meaning |
|---|---|
| Identity | agent:{agentId} — addressable, auditable, tenant-scoped |
| Role | The organizational role and responsibility it represents (e.g., Renewal Agent, Support Triage Agent) |
| Authority | What it may do in a given Work context — bounded by policy and governance verdict |
| Memory | Agent Continuity — working state that serves Business Context, not a private data lake |
| Capabilities | The governed subset of the Capability Fabric available to this agent in this Work |
| Work | The Work or subtask it owns, with state, evidence, and handoff linkage |
| Business Context | The Spine projection it is authorized to operate on |
Agent Work Lifecycle
An agent takes ownership of a Work subtask (often handed from Twin or another agent) with explicit identity and authority.
Hydrates Active Context Bundle — Business Context slice + Work + evidence + authority + agent continuity. No re-briefing needed.
Observes current state, orients on business objective, plans capability sequence — inside the AI Workbench, not as exposed buttons.
Resolves provider-neutral capabilities, submits plan to governance for verdict. Denied → revise. Approved → proceed.
Hermes executes. Outcome is verified vs expected, evidence captured, and reconciliation decides what updates the Spine. Projection invalidates; workbenches rehydrate.
Leaves evidence on Shared Work. Next agent or human continues from updated Spine truth — no transcript pasting.
Boundaries
- An agent never bypasses governance or acts outside its authorized Business Context projection.
- An agent never promotes working inference to organizational truth without the governed write → verify → reconcile path.
- Agent coordination is via Shared Work + Shared Business Context + Evidence, not via describing the organization repeatedly in chat.
Where to Go Next
<CardGroup cols={2}>How multiple agents share Business Context and Work without re-explaining everything.
How 'may it be done?' is decided for this agent in this Work.
How authorized agent plans are executed.