Unclassified

Agent Fabric

The orchestration layer for specialized agents — one user-facing Twin, invisible backstage workers, governed handoff.

Agent Fabric is the orchestration layer for all agents in SpineWorkspace. It follows one master doctrine:

One Workspace. One Spine. One shared Agent Home. One Twin. Many invisible capabilities.

Users work with their business. Twin carries continuity. Agents operate backstage. The system surfaces outcomes, evidence, decisions, and required human attention — not agent machinery.

Core Principle: Single Face to the User

The human user has one user-facing relationship: their Twin.

Twin is the sole user-facing orchestrator. Specialist agents operate inside the shared Agent Home and may be orchestrated by Twin, workflows, events, or runtime policy. Their outputs surface to users through Twin or the appropriate Workbench projection — not as competing agents.

ONE SPINEWORKSPACE
        │
        ▼
ONE SHARED AGENT HOME
        │
        ▼
      TWIN
single continuity-bearing,
user-facing orchestrator
        │
        ├── specialists (Triage, Schema-AI, Coworker)
        ├── skills & playbooks
        ├── workflows & state machines
        ├── models (dynamic cost/latency routing)
        ├── tools / MCP connector pool
        ├── jobs & normalizers
        ├── schedules & crons
        └── runtime workers

            invisible (engine room)

USER
 │
 ▼
Work → Awareness → Decide → Act → Continue

Agent Home

Agent Home is the runtime and working state for all agents. It is the shared engine room — not visible to users, not a source of truth, not a replacement for Spine.

Contains:

Agent session state
Working memory and scratchpads
Task queues and assignment state
Model routing and cost optimization
Tool/MCP connector pool
Scheduled jobs and crons

Rules:

Agent Home is ephemeral — no authority
Agent Home state does not persist as canonical truth
Agent Home is scoped to tenant and workspace

Specialized Worker Fleet

Specialized Capability	Role & Responsibility	Surfacing Rule
Triage Agent	Scans webhooks, emails, tickets, telemetry. Categorizes severity and domain.	Surfaces to Twin/Workbench. Never pings users directly.
Schema Agent	Ingests multi-source data, normalizes into canonical Spine schemas.	Reconciles backstage; never prompts user for mapping chores.
Coworker Agent	Domain-specific background reasoning (churn analysis, reconciliation).	Submits structured drafts for Twin synthesis.
Adaptive Auto-Tuner	Monitors latency, cost, accuracy; dynamically switches model routing.	Autonomous runtime policy. Optimizes invisibly.
Hermes Executor	Dispatches authenticated API mutations after human sign-off.	Logs execution proofs to Fortress Ledger.

Agent Types

From the canonical ontology:

Type	Purpose	User-Facing	Authority
Twin	User's AI work partner	Yes	Bounded by role
Coworker	Context gathering and support	Yes	Read-only context
Triage	Classification and routing	No	Operational state only
Domain Agent	Domain-specific expertise	No	Scoped context
Specialist Agent	Bounded task execution	No	Scoped context
Workflow Agent	Multi-step orchestration	No	Governed execution
System Agent	Platform operations	No	Infrastructure scope
External Agent	Third-party integration	No	Connector scope

Agent Boundaries

Agents:

Receive scoped context
Return results/evidence
Do not independently own canonical Work or Organizational Memory
Operate within the AI Operational Sandbox
Execute only after governance approval

Agents do NOT:

Own canonical Spine truth
Bypass governance
Mutate organizational memory directly
Act without authorization
Replace human judgment for consequential decisions

ACP vs MCP

Protocol	Purpose	Scope
MCP	AI ↔ Tool / Resource / Capability	Tool invocation
ACP	Agent ↔ Agent / Agent ↔ Twin	Work/context handoff

Neither grants authority. Both preserve tenant, identity, work, context, purpose, provenance, correlation.

Four Golden Rules

Sole User-Facing Presenter: Twin is the only agent that speaks to the human user.
Context & Continuity Stewardship: Twin preserves cross-session continuity, role awareness, working context.
Decoupled Internal Execution: Internal workers run asynchronously without blocking Twin or polluting user focus.
Governed Human Sign-Off: Sensitive actions encapsulated into Proposal Cards for human approval.

Twin — the user-facing orchestrator
Capability Fabric — what agents can execute
Governance — the policy boundary agents respect
Reconciliation — how execution results are verified