{"slug":"architecture/CAPABILITY_FABRIC","title":"Capability Fabric","section":"architecture","sectionLabel":"Architecture","status":"unclassified","type":"document","path":"architecture/CAPABILITY_FABRIC.mdx","bytes":4081,"description":"The governed execution language between intelligence and operational action — discover, select, validate, execute.","updated":null,"source":"integratewise/Integratewise-docs-wiki","content":"\nCapability Fabric is the governed execution language between intelligence and operational action. It answers the question:\n\n> **WHAT CAN BE DONE?**\n\n## What Capability Fabric Does\n\n- Discovers capabilities from the registry\n- Selects capabilities by policy and context\n- Validates parameters against capability contracts\n- Enforces governance gates before execution\n- Routes execution to the correct runtime implementation\n- Returns the result as canonical evidence\n\n## What Capability Fabric Does NOT Do\n\n- Own canonical truth\n- Decide outcomes alone\n- Bypass governance\n- Provide direct operational memory\n\n```text\nINTELLIGENCE\n      ↓\nCAPABILITY SELECTION\n      ↓\nPOLICY / GOVERNANCE\n      ↓\nEXECUTION\n      ↓\nRESULT\n      ↓\nCANONICAL CONTINUITY\n```\n\n## Capability Registry\n\nThe capability registry is the catalog of all available capabilities. Each capability is defined by:\n\n| Field | Description |\n|-------|-------------|\n| **key** | Unique identifier (e.g., `retry_payment`, `create_task`) |\n| **plane** | Which architectural plane it operates in |\n| **inputs** | Expected parameters and types |\n| **outputs** | Return values and types |\n| **required_scopes** | Authorization scopes needed |\n| **governance_policy** | What governance is required |\n| **surface_compatibility** | Which surfaces can invoke this capability |\n\n## Governance Contract\n\nEvery capability execution path must include a governed boundary:\n\n```text\nCapability selection\n      ↓\nPolicy evaluation\n      ↓\nGovernance gate if required by policy\n      ↓\nExecution contract accepted\n      ↓\nCapability execution\n      ↓\nCanonical result path\n```\n\n**Rule:** Capability Fabric must not bypass governance.\n\n## Implementation Types\n\nCapability implementation can be:\n\n| Type | Description |\n|------|-------------|\n| **API** | Direct REST/GraphQL call to external service |\n| **MCP** | Tool invocation via Model Context Protocol |\n| **Spine Tool** | Internal Spine operation |\n| **Workflow** | Multi-step orchestrated flow |\n| **Agent** | AI agent execution |\n| **ADK** | Agent Development Kit mechanism |\n| **Provider** | External provider mechanism |\n\n## Discovery and Marketplace\n\nThree related marketplaces support capability discovery:\n\n| Marketplace | Purpose |\n|-------------|---------|\n| **Integration Marketplace** | External system connectors |\n| **Plugin Marketplace** | Workbench extensions |\n| **MCP Marketplace** | AI tool capabilities |\n\n**Lifecycle:**\n```text\nDiscover → Validate → Install/Connect → Authorize → Register Capability → Govern → Execute\n```\n\n**Rules:**\n- Marketplace is capability discovery/distribution\n- Marketplace is not canonical truth\n- Plugin installed ≠ permission\n- MCP available ≠ authority\n\n## Capability Examples\n\n| Capability | Description | Governance |\n|-----------|-------------|------------|\n| `retry_payment` | Retry a failed payment | Approval required |\n| `create_task` | Create a new work item | Policy check |\n| `send_customer_message` | Send message to customer | Approval required |\n| `update_account` | Update account record | Role check |\n| `create_invoice` | Generate an invoice | Approval required |\n| `schedule_meeting` | Schedule a meeting | Policy check |\n| `reconcile_payment` | Match payment to invoice | Role check |\n\n## Capability and Execution\n\nCapability is separate from execution. The flow is:\n\n```text\nCAPABILITY (what can be done)\n      ↓\nEXECUTION FABRIC (how it is done)\n      ↓\nGOVERNANCE (may it be done)\n      ↓\nRESULT (what happened)\n      ↓\nRECONCILIATION (is it correct)\n```\n\n## Related\n\n- [Execution Fabric](/architecture/EXECUTION_FABRIC) — how capabilities are executed\n- [Governance](/architecture/GOVERNANCE) — the policy boundary\n- [Twin](/architecture/TWIN) — who invokes capabilities\n- [Agent Fabric](/architecture/AGENT_FABRIC) — the agent fleet that uses capabilities\n"}