{"slug":"business/v2-ecosystem","title":"§E — Ecosystem Architecture & SDK Specification","section":"business","sectionLabel":"Business","status":"unclassified","type":"document","path":"business/v2-ecosystem.mdx","bytes":85106,"description":null,"updated":null,"source":"integratewise/Integratewise-docs-wiki","content":"\n> **Status: SUPERSEDED DEEP-DIVE.** The old product name, layer model and\n> permanent Twin-panel examples are historical only. Current Capability,\n> Workbench and Silent Twin rules live in\n> [`ARCHITECTURE_LOCK.md`](./ARCHITECTURE_LOCK.md) and\n> [`TWIN_BEHAVIOR.md`](./TWIN_BEHAVIOR.md).\n\n## IntegrateWise Continuity Bridge v2.0 — Capability Fabric, Knowledge Protocol, A2A, SDK Variants, Composition Patterns\n\n**Status:** DRAFT v2.0 — Stage 1/2 Deep-Dive \n**Domain Owner:** Ecosystem Architecture & SDK \n**Supersedes:** v1.0.1 §5 (MCP Pool), §9 (Discovery), §12.4 (A2A), §18 (Registry), §26 (Template) where conflicting. \n**Date:** 2026-07-02\n\n---\n\n## E.1 Doctrine — What the Ecosystem Layer Is\n\n> **The Ecosystem is the capability fabric.** It is not a separate system. It is the emergent property of the Continuity Bridge when tools, agents, humans, and memory are allowed to communicate through the Spine — governed, audited, and scoped.\n\nThe Ecosystem Layer sits conceptually across the **Capability** and **Continuity** layers of the 6-layer stack. It is the substrate that enables:\n\n1. **Capability Fabric Model** — How named business capabilities are discovered, composed, and executed across the mesh.\n2. **Knowledge Sharing Protocol (KSP)** — How memory, context, and evidence move between agents, tools, and humans without leaking tenant boundaries.\n3. **Agent-to-Agent Communication (A2A)** — How autonomous agents coordinate through the Spine (not peer-to-peer), governed by the Twin ambient reasoning layer.\n4. **SDK Variants** — Five consumer-facing SDK packages that expose the same contract at different fidelity levels.\n5. **Ecosystem App Composition Patterns** — How third-party developers build on the Bridge without importing internal topology.\n6. **Template Starter** — The canonical `create-iw-app` bootstrap for ecosystem participants.\n\n**Key Invariant:** The Ecosystem never bypasses the Spine. All communication between ecosystem participants is mediated by the Spine, gated by `tenant_id`, and audited to `spine_audit_log`.\n\n---\n\n## E.2 The Capability Fabric Model\n\n### E.2.1 Conceptual Architecture\n\n```\n┌─────────────────────────────────────────────────────────────────────────────┐\n│ ECOSYSTEM CAPABILITY FABRIC │\n│ │\n│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │\n│ │ Capability │◄──►│ Capability │◄──►│ Capability │◄──►│ Capability │ │\n│ │ Registry │ │ Router │ │ Sandbox │ │ Monitor │ │\n│ │(agent-registry) │ (hermes) │ │(iw-agent- │ │ (telemetry) │ │\n│ └──────┬──────┘ └──────┬──────┘ │ runtime) │ └─────────────┘ │\n│ │ │ └─────────────┘ │\n│ ▼ ▼ │\n│ ┌─────────────────────────────────────────────────────────────────────┐ │\n│ │ SPINE — The Single Source of Truth │ │\n│ │ (entities, relationships, memory, audit, continuity) │ │\n│ └─────────────────────────────────────────────────────────────────────┘ │\n│ ▲ │\n│ ┌─────────────┐ ┌───────┴───────┐ ┌─────────────┐ ┌──────────┐ │\n│ │ Inbound │ │ Twin │ │ Outbound │ │ Shared │ │\n│ │ MCP Pool │◄──►│ (ambient) │◄──►│ MCP Pool │◄──►│ Memory │ │\n│ │(mcp-connector) │(twin-orchestrator) │ (act + │ │(continuity│ │\n│ └─────────────┘ └─────────────┘ │ adapters) │ └──────────┘ │\n│ └─────────────┘ │\n└─────────────────────────────────────────────────────────────────────────────┘\n ▲\n Consumers ────────────┴───────────── SDKs, AI Assistants, Web Workbench\n```\n\n### E.2.2 Capability Taxonomy\n\nEvery capability in the ecosystem is a **named, versioned, discoverable action** with a strict contract:\n\n```typescript\n// packages/sdk/src/types/capability.ts\n\n/**\n * Capability — the atomic unit of the Capability Fabric.\n * Every capability is: named, versioned, scoped, gated, and audited.\n */\nexport interface Capability {\n /** Canonical capability ID: `{domain}.{resource}.{action}` */\n id: string; // e.g., \"salesforce.opportunity.create\"\n\n /** Semantic version of the capability contract */\n version: string; // e.g., \"2.1.0\"\n\n /** Human-readable name */\n name: string;\n\n /** Markdown description for Discovery */\n description: string;\n\n /** The provider that owns this capability */\n provider: string; // e.g., \"salesforce\", \"slack\", \"iw-native\"\n\n /** Capability tier: L1=read, L2=insight, L3=execute, L4=autonomous */\n level: 1 | 2 | 3 | 4;\n\n /** Required scopes to invoke */\n scopes: string[]; // e.g., [\"read:opportunities\", \"write:opportunities\"]\n\n /** Input JSON Schema (for validation + UI generation) */\n inputSchema: JSONSchema7;\n\n /** Output JSON Schema */\n outputSchema: JSONSchema7;\n\n /** Governance gate required? */\n governanceGate: boolean; // true for L3/L4, false for L1/L2\n\n /** Minimum confidence threshold for auto-execution */\n autoExecuteThreshold: number; // 0.85 default for L3, 1.0 (never auto) for L4\n\n /** Whether this capability supports streaming (SSE) responses */\n streaming: boolean;\n\n /** Estimated cost in tokens / compute units */\n costEstimate: CostEstimate;\n\n /** Deprecation status */\n deprecation?: {\n deprecatedAt: string; // ISO date\n sunsetAt: string; // ISO date\n replacement?: string; // capability ID of replacement\n };\n}\n\nexport interface CostEstimate {\n /** Token estimate for LLM-based capabilities */\n tokens?: { input: number; output: number };\n\n /** External API call cost estimate (USD) */\n apiCalls?: { count: number; estimatedCost: number };\n\n /** Compute time estimate (ms) */\n computeMs: number;\n}\n\n/** Capability invocation request — what consumers send */\nexport interface CapabilityInvokeRequest {\n /** The capability to invoke */\n capability: string;\n\n /** Tenant context (injected by Gateway, never from client) */\n tenantId: string;\n\n /** User context (injected by Gateway from JWT) */\n userId: string;\n\n /** Correlation ID for distributed tracing */\n correlationId: string;\n\n /** Invocation payload — validated against capability.inputSchema */\n payload: unknown;\n\n /** Execution preferences */\n options?: {\n /** Prefer streaming response? */\n streaming?: boolean;\n\n /** Maximum wait time (ms) before async handoff */\n timeoutMs?: number;\n\n /** Whether to compound result into memory */\n compoundMemory?: boolean;\n\n /** Required confidence threshold override */\n confidenceThreshold?: number;\n };\n}\n\n/** Capability invocation response */\nexport interface CapabilityInvokeResponse {\n correlationId: string;\n status: \"success\" | \"pending\" | \"rejected\" | \"error\" | \"async\";\n result?: unknown;\n proposalId?: string; // Present if governance-gated\n confidence?: number; // Governance confidence score\n auditLogId: string; // Reference to spine_audit_log entry\n executionTimeMs: number;\n}\n```\n\n### E.2.3 Capability Discovery & Routing\n\n```\n┌─────────────────────────────────────────────────────────────────────────────┐\n│ CAPABILITY DISCOVERY & ROUTING FLOW │\n│ │\n│ Consumer Gateway Discovery │\n│ │ │ Cache │\n│ │ GET /api/v1/discovery │ │ │\n│ │ ─────────────────────────────► │ │ │\n│ │ │ 1. Extract tenantId from JWT │ │\n│ │ │ 2. Query KV cache (5-min TTL) │ │\n│ │ │ │ │\n│ │ │ ◄──── cached Discovery? ───────►│ │\n│ │ │ │ │\n│ │ │ [Cache MISS] │ │\n│ │ │ │ │ │\n│ │ │ ▼ │ │\n│ │ │ 3. Fetch from agent-registry │ │\n│ │ │ (service binding) │ │\n│ │ │ │ │\n│ │ │ 4. agent-registry queries: │ │\n│ │ │ - capability_registry (D1) │ │\n│ │ │ - endpoint_registry (D1) │ │\n│ │ │ - connector status (KV) │ │\n│ │ │ │ │\n│ │ │ 5. Filter by tenant plan + RBAC │ │\n│ │ │ │ │\n│ │ │ 6. Write to KV cache │ │\n│ │ │ │ │\n│ │ DiscoveryResponse │ │ │\n│ │ ◄───────────────────────────── │ │ │\n│ │ │ │ │\n└─────────────────────────────────────────────────────────────────────────────┘\n```\n\n**Discovery Response Contract:**\n\n```typescript\n// packages/sdk/src/types/discovery.ts\n\nexport interface DiscoveryResponse {\n /** Identity context */\n identity: {\n userId: string;\n tenantId: string;\n orgId: string;\n role: string;\n scopes: string[];\n };\n\n /** Workspace configuration */\n workspace: {\n name: string;\n plan: \"free\" | \"pro\" | \"enterprise\";\n connectedProviders: string[];\n limits: WorkspaceLimits;\n };\n\n /** Available projections (8 lenses) */\n projections: Projection[];\n\n /** Available capabilities — FILTERED by tenant plan + RBAC */\n capabilities: CapabilitySummary[];\n\n /** Available agents in the ecosystem */\n agents: AgentSummary[];\n\n /** Knowledge sources available */\n knowledgeSources: KnowledgeSource[];\n\n /** Navigation structure for the workbench */\n navigation: NavNode[];\n\n /** Personalization */\n personalization: {\n recentCapabilities: string[];\n pinnedProjections: string[];\n preferences: Record;\n };\n\n /** SDK version compatibility */\n sdkVersion: string;\n\n /** Bridge version */\n bridgeVersion: string;\n}\n\nexport interface CapabilitySummary {\n id: string;\n name: string;\n description: string;\n level: 1 | 2 | 3 | 4;\n provider: string;\n category: string;\n scopes: string[];\n streaming: boolean;\n /** Whether currently available (provider connected + healthy) */\n available: boolean;\n}\n\nexport interface AgentSummary {\n id: string;\n name: string;\n description: string;\n capabilities: string[];\n status: \"active\" | \"idle\" | \"maintenance\";\n}\n\nexport interface KnowledgeSource {\n id: string;\n name: string;\n type: \"spine\" | \"memory\" | \"document\" | \"connector\";\n entityTypes: string[];\n}\n```\n\n### E.2.4 Capability Execution Flow\n\n```\nConsumer → Gateway → hermes (queue) → twin-orchestrator (OODA) → govern (gate) → act (execute)\n │\n ▼\n ┌─────────────────┐\n │ SPINE READ │\n │ Entity360 + │\n │ Memory + RAG │\n └─────────────────┘\n```\n\n**Detailed Execution Flow:**\n\n| Step | Service | Queue/Binding | Action | Security Boundary |\n| ---- | ------------------- | --------------------------------- | ---------------------------------------------------------------- | ----------------------- |\n| 1 | `gateway` | N/A | Validate JWT, extract `tenant_id`, assemble `correlationId` | Identity Layer |\n| 2 | `gateway` | `HERMES` (service binding) | Enqueue `CapabilityInvokeRequest` to Hermes | Ingress → Capability |\n| 3 | `hermes` | `ACT_QUEUE` | DO acquires message, checks idempotency key | Capability Layer |\n| 4 | `hermes` | `TWIN_ORCHESTRATOR` (binding) | If L2/L3/L4, route to Twin for OODA | Capability → Continuity |\n| 5 | `twin-orchestrator` | D1 + KV + Vectorize | Orient: assemble Entity360 + Memory + Knowledge | Continuity Layer |\n| 6 | `twin-orchestrator` | `THINK` (binding) | Decide: LLM reasoning, proposal generation | Continuity Layer |\n| 7 | `twin-orchestrator` | `GOVERN` (binding) | Confidence scoring, proposal emission | Governance Layer |\n| 8 | `govern` | D1 `action_proposals` | Gate: ≥0.85 auto, 0.70-0.85 HITL, <0.70 discard | Governance Layer |\n| 9 | `govern` | `ACT_QUEUE` | Approved → enqueue to act | Governance → Provider |\n| 10 | `act` | D1 + outbound adapters | Execute: adapter selection, credential wall, Entity360 injection | Provider Fabric |\n| 11 | `act` | `PIPELINE_QUEUE` | Writeback: result → Pipeline → Spine | Provider → Spine |\n| 12 | `pipeline` | D1 `entities` + `spine_audit_log` | Commit: ONLY writer to Spine | Spine Integrity |\n| 13 | `continuity` | D1 `memory` + KV | Remember: compound into memory tiers | Memory Layer |\n\n---\n\n## E.3 Knowledge Sharing Protocol (KSP)\n\n### E.3.1 Protocol Overview\n\n> **KSP is the contract for how knowledge moves between ecosystem participants.** It is not a separate network protocol. It is a schema and governance contract over the Spine.\n\n```\n┌─────────────────────────────────────────────────────────────────────────────┐\n│ KNOWLEDGE SHARING PROTOCOL (KSP) │\n│ │\n│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │\n│ │ Agent A │◄───►│ SPINE │◄───►│ Agent B │◄───►│ Human │ │\n│ │ (twin-A) │ │ (shared) │ │ (twin-B) │ │ (desk) │ │\n│ └────┬─────┘ └────┬─────┘ └────┬─────┘ └────┬─────┘ │\n│ │ │ │ │ │\n│ │ KSP Packet │ KSP Packet │ KSP Packet │ │\n│ │ (scoped) │ (scoped) │ (scoped) │ │\n│ │ │ │ │ │\n│ ┌────▼────────────────▼────────────────▼────────────────▼─────┐ │\n│ │ KSP GOVERNANCE LAYER (govern) │ │\n│ │ • tenant_id isolation (HARD) │ │\n│ │ • scope validation (read/write/admin) │ │\n│ │ • confidence scoring (0.70/0.85) │ │\n│ │ • audit trail (spine_audit_log) │ │\n│ │ • retention policy (D1 + R2) │ │\n│ └─────────────────────────────────────────────────────────────┘ │\n│ │\n└─────────────────────────────────────────────────────────────────────────────┘\n```\n\n### E.3.2 KSP Packet Schema\n\n```typescript\n// packages/sdk/src/types/ksp.ts\n\n/**\n * Knowledge Sharing Protocol Packet — the unit of knowledge transfer.\n * Every packet is: typed, scoped, attributed, timed, and governed.\n */\nexport interface KSPPacket {\n /** Unique packet ID (UUID v7 for time-sortability) */\n id: string;\n\n /** Protocol version */\n version: \"ksp/2.0\";\n\n /** Tenant isolation boundary — HARD GATE */\n tenantId: string;\n\n /** Source of the knowledge */\n source: KSPSource;\n\n /** Destination of the knowledge */\n destination: KSPDestination;\n\n /** The knowledge payload */\n payload: KSPPayload;\n\n /** Temporal context */\n timing: KSPTiming;\n\n /** Governance metadata */\n governance: KSPGovernance;\n\n /** Audit reference */\n audit: {\n logId: string;\n spineAuditLogId: string;\n };\n}\n\nexport interface KSPSource {\n /** Source type */\n type: \"agent\" | \"tool\" | \"human\" | \"twin\" | \"system\" | \"connector\";\n\n /** Source ID (agent ID, user ID, tool ID) */\n id: string;\n\n /** Source name (for display) */\n name: string;\n\n /** Capability that produced this knowledge (if applicable) */\n capabilityId?: string;\n\n /** Provider that produced this knowledge (if applicable) */\n providerId?: string;\n}\n\nexport interface KSPDestination {\n /** Target type */\n type: \"broadcast\" | \"agent\" | \"tool\" | \"human\" | \"twin\" | \"memory\" | \"spine\";\n\n /** Target ID (for directed packets) */\n id?: string;\n\n /** Scope filter — who can see this packet */\n visibility: \"personal\" | \"work\" | \"organization\" | \"ai\";\n}\n\nexport interface KSPPayload {\n /** Payload type discriminator */\n type:\n | \"observation\"\n | \"orientation\"\n | \"decision\"\n | \"action_result\"\n | \"memory_fragment\"\n | \"signal\"\n | \"proposal\"\n | \"evidence\";\n\n /** The actual content — shape depends on type */\n content: unknown;\n\n /** Semantic embedding (Vectorize) for similarity search */\n embedding?: number[];\n\n /** Confidence score (0.0–1.0) for memory triage */\n confidence: number;\n\n /** References to related Spine entities */\n entityRefs?: string[]; // SSOC UUIDs\n\n /** References to related packets (lineage) */\n parentPacketIds?: string[];\n}\n\nexport interface KSPTiming {\n /** Packet creation time */\n createdAt: string; // ISO 8601\n\n /** Valid until (for time-bounded knowledge) */\n expiresAt?: string;\n\n /** Time-to-live in seconds (for cacheable knowledge) */\n ttlSeconds?: number;\n}\n\nexport interface KSPGovernance {\n /** Minimum role required to access */\n minRole: string;\n\n /** Required scopes */\n requiredScopes: string[];\n\n /** Whether this packet requires governance approval before distribution */\n approvalRequired: boolean;\n\n /** Classification level */\n classification: \"public\" | \"internal\" | \"confidential\" | \"restricted\";\n\n /** Data residency requirements */\n residency?: string; // e.g., \"us-east\", \"eu-west\"\n}\n```\n\n### E.3.3 KSP Flow — Agent Shares Insight with Organization\n\n```\n┌─────────────────────────────────────────────────────────────────────────────┐\n│ KSP FLOW: Agent A Shares Insight → Organization Memory │\n│ │\n│ Agent A (twin-orchestrator-A) │\n│ │ │\n│ │ 1. Generates insight (OODA Decide phase) │\n│ │ confidence: 0.92 │\n│ │ │\n│ ▼ │\n│ ┌─────────────────────────────────────────────────────────────────────┐ │\n│ │ 2. Assemble KSP Packet │ │\n│ │ source: { type: \"twin\", id: \"twin-A\", name: \"Sales Twin\" } │ │\n│ │ destination: { type: \"memory\", visibility: \"organization\" } │ │\n│ │ payload: { type: \"memory_fragment\", confidence: 0.92, ... } │ │\n│ │ governance: { minRole: \"member\", approvalRequired: false } │ │\n│ └─────────────────────────────────────────────────────────────────────┘ │\n│ │ │\n│ │ POST /api/v1/knowledge/share │\n│ ▼ │\n│ Gateway — validates JWT, extracts tenant_id │\n│ │ │\n│ │ 3. Route to knowledge service (service binding) │\n│ ▼ │\n│ knowledge — semantic analysis, embedding generation │\n│ │ │\n│ │ 4. Route to continuity (service binding) │\n│ ▼ │\n│ continuity — Memory Triage │\n│ │ confidence ≥ 0.85 → ACTIVE memory (D1 + KV + Vectorize) │\n│ │ 0.70–0.85 → STAGING memory (D1 only) │\n│ │ < 0.70 → DISCARD (R2 cold storage) │\n│ │ │\n│ │ 5. Write to spine_audit_log │\n│ ▼ │\n│ pipeline — commit audit entry │\n│ │ │\n│ │ 6. Broadcast to subscribers │\n│ ▼ │\n│ hermes — enqueue to notification queue │\n│ │ │\n│ │ SSE push to connected workbench clients │\n│ │ WebSocket update to Twin Workbench │\n│ ▼ │\n│ Organization Memory Updated │\n└─────────────────────────────────────────────────────────────────────────────┘\n```\n\n### E.3.4 KSP Security Boundaries\n\n| Boundary | Enforcement | Failure Mode |\n| --------------------- | -------------------------------------------------- | --------------------- |\n| **Tenant Isolation** | `WHERE tenant_id = ?` on every KSP query | 403 fail-loud |\n| **Visibility Scope** | `personal` < `work` < `organization` < `ai` | Filter silently |\n| **Role Gate** | `minRole` checked against RBAC in `tenants` | 403 if insufficient |\n| **Scope Gate** | `requiredScopes` checked against JWT `scope` claim | 403 if missing |\n| **Classification** | `restricted` → requires `admin` role + approval | Quarantine to staging |\n| **Residency** | Packet rejected if target region mismatch | 422 with reason |\n| **Approval Required** | `approvalRequired: true` → routed to `govern` | HITL queue |\n\n---\n\n## E.4 Agent-to-Agent Communication (A2A)\n\n### E.4.1 A2A Architecture — Spine as Shared Blackboard\n\n> **A2A in v2.0 is NOT peer-to-peer.** Agents coordinate through the Spine as a shared blackboard, mediated by the ambient Twin layer. This prevents circular dependencies, ensures auditability, and maintains tenant isolation.\n\n```\n┌─────────────────────────────────────────────────────────────────────────────┐\n│ A2A: SPINE-BASED COORDINATION (NOT PEER-TO-PEER) │\n│ │\n│ │\n│ Agent A ──────┐ │\n│ (Sales Twin) │ │\n│ │ │\n│ │ ┌──────────────────────────────────────────┐ │\n│ │ │ SPINE BLACKBOARD │ │\n│ ├───►│ │ │\n│ │ │ ┌─────────────┐ ┌─────────────┐ │ │\n│ Agent B ──────┤ │ │ Intent A │ │ Intent B │ │ │\n│ (CS Twin) │ │ │ (proposal) │ │ (proposal) │ │ │\n│ │ │ └──────┬──────┘ └──────┬──────┘ │ │\n│ │ │ │ │ │ │\n│ │ │ ▼ ▼ │ │\n│ │ │ ┌─────────────────────────────────┐ │ │\n│ ├───►│ │ GOVERN (arbitration) │ │ │\n│ │ │ │ • Detect conflicts │ │ │\n│ Agent C ──────┤ │ │ • Score joint confidence │ │ │\n│ (Finance Twin) │ │ │ • Enforce priority rules │ │ │\n│ │ │ │ • Escalate to human if needed │ │ │\n│ │ │ └─────────────────────────────────┘ │ │\n│ │ │ │ │ │ │\n│ │ │ ▼ ▼ │ │\n│ │ │ ┌─────────────┐ ┌─────────────┐ │ │\n│ └───►│ │ Action A │ │ Action B │ │ │\n│ │ │ (approved) │ │ (approved) │ │ │\n│ │ └─────────────┘ └─────────────┘ │ │\n│ │ │ │\n│ └──────────────────────────────────────────┘ │\n│ ▲ │\n│ │ │\n│ ┌───────┴───────┐ │\n│ │ TWIN LAYER │ │\n│ │ (ambient) │ │\n│ │ • OODA cycle │ │\n│ │ • Context asm │ │\n│ │ • Reasoning │ │\n│ └───────────────┘ │\n│ │\n│ RULE: Agents never call each other directly. They read/write the Spine. │\n│ The Twin layer observes and reasons across agent outputs. │\n│ │\n└─────────────────────────────────────────────────────────────────────────────┘\n```\n\n### E.4.2 A2A Intent Schema\n\n```typescript\n// packages/sdk/src/types/a2a.ts\n\n/**\n * A2A Intent — an agent's declared intention published to the Spine.\n * Other agents read intents and may respond with complementary or conflicting intents.\n */\nexport interface A2AIntent {\n /** Unique intent ID */\n id: string;\n\n /** The agent that published this intent */\n agentId: string;\n\n /** Agent type */\n agentType: \"twin\" | \"autonomous\" | \"workflow\" | \"human\";\n\n /** Tenant isolation */\n tenantId: string;\n\n /** Intent classification */\n intentType: A2AIntentType;\n\n /** The target capability or entity */\n target: {\n type: \"capability\" | \"entity\" | \"memory\" | \"projection\";\n id: string;\n };\n\n /** What the agent wants to do */\n action: {\n verb: \"propose\" | \"request\" | \"inform\" | \"query\" | \"warn\" | \"block\";\n description: string;\n payload: unknown;\n };\n\n /** Confidence in this intent (0.0–1.0) */\n confidence: number;\n\n /** Priority relative to other intents */\n priority: \"critical\" | \"high\" | \"normal\" | \"low\";\n\n /** Time bounds */\n timing: {\n publishedAt: string;\n validUntil?: string;\n executeAfter?: string; // Defer execution until this time\n };\n\n /** Dependencies — other intents that must resolve first */\n dependencies?: string[]; // Intent IDs\n\n /** Response to another intent (for threading) */\n inReplyTo?: string;\n\n /** Governance status */\n governance: {\n status: \"pending\" | \"approved\" | \"rejected\" | \"escalated\";\n proposalId?: string;\n approvedBy?: string;\n approvedAt?: string;\n };\n}\n\nexport type A2AIntentType =\n | \"cross-sell\" // Sales agent → CS agent: opportunity signal\n | \"churn-risk\" // CS agent → Sales/Finance: retention alert\n | \"data-request\" // Any agent → Spine: needs context\n | \"action-proposal\" // Any agent → Govern: proposes action\n | \"memory-share\" // Any agent → Memory: shares insight\n | \"conflict\" // Agent → Agent: detects conflicting intent\n | \"handoff\" // Agent → Human: needs human intervention\n | \"sync-request\"; // Agent → Connector: trigger sync\n\n/**\n * A2A Arbitration Result — produced by the Twin layer when conflicting intents detected.\n */\nexport interface A2AArbitration {\n id: string;\n tenantId: string;\n conflictType: \"resource\" | \"priority\" | \"dependency\" | \"policy\";\n conflictingIntents: string[];\n resolution: \"merge\" | \"sequence\" | \"reject\" | \"escalate\";\n mergedIntent?: A2AIntent;\n sequence?: string[]; // Ordered intent IDs\n reason: string;\n arbitratedBy: string; // Twin instance ID\n arbitratedAt: string;\n}\n```\n\n### E.4.3 A2A Flow — Cross-Sell Signal from Sales Twin to CS Twin\n\n```\n┌─────────────────────────────────────────────────────────────────────────────┐\n│ A2A FLOW: Sales Twin Detects Upsell → CS Twin Takes Action │\n│ │\n│ SALES TWIN (twin-orchestrator-sales) │\n│ │ │\n│ │ 1. OBSERVE: Account \"Acme Corp\" shows expansion signals │\n│ │ (new team members, API usage spike) │\n│ │ │\n│ │ 2. ORIENT: Entity360 shows CS health score = 0.91 │\n│ │ Memory: recent positive NPS, no open tickets │\n│ │ │\n│ │ 3. DECIDE: Publish \"cross-sell\" intent to Spine │\n│ ▼ │\n│ ┌─────────────────────────────────────────────────────────────────────┐ │\n│ │ A2AIntent: │ │\n│ │ agentId: \"twin-sales\" │ │\n│ │ intentType: \"cross-sell\" │ │\n│ │ target: { type: \"entity\", id: \"account:acme-corp\" } │ │\n│ │ action: { verb: \"propose\", description: \"Upsell to Enterprise\", │ │\n│ │ payload: { expansionSignals: [...] } } │ │\n│ │ confidence: 0.89 │ │\n│ │ priority: \"high\" │ │\n│ └─────────────────────────────────────────────────────────────────────┘ │\n│ │ │\n│ │ POST via mcp-connector → Gateway → pipeline │\n│ ▼ │\n│ SPINE — intent stored in `a2a_intents` table (D1) │\n│ │ │\n│ │ 4. TWIN LAYER (ambient) observes new intent │\n│ │ twin-orchestrator reads all unprocessed intents for tenant │\n│ ▼ │\n│ CS TWIN (twin-orchestrator-cs) │\n│ │ │\n│ │ 5. Reads intent from Spine (WHERE tenant_id = ? AND agentId != self) │\n│ │ │\n│ │ 6. ORIENT: Fetch Account context, recent tickets, NPS │\n│ │ │\n│ │ 7. DECIDE: Confidence = 0.87, agrees with upsell │\n│ │ Publishes response intent: \"action-proposal\" to govern │\n│ ▼ │\n│ GOVERN — joint confidence = 0.88 (0.89 × 0.87 weighted) │\n│ │ ≥ 0.85 → AUTO-APPROVE │\n│ ▼ │\n│ ACT — execute \"salesforce.opportunity.create\" + \"slack.notify-cs-channel\" │\n│ │ │\n│ ▼ │\n│ PIPELINE — writeback to Spine │\n│ CONTINUITY — compound into Organization Memory │\n│ │\n│ RESULT: CS team notified in Slack with full context. No manual routing. │\n│ Audit trail: spine_audit_log entries for every step. │\n└─────────────────────────────────────────────────────────────────────────────┘\n```\n\n### E.4.4 A2A Security Model\n\n| Threat | Mitigation | Enforcement Point |\n| --------------------------- | ------------------------------------------------ | ---------------------------- |\n| Agent impersonation | JWT signing + `agent_id` registry | `gateway` + `agent-registry` |\n| Cross-tenant intent leakage | `WHERE tenant_id = ?` on ALL intent queries | All services |\n| Intent flooding | Rate-limit per agent: 100 intents/min | `gateway` + `hermes` |\n| Circular dependencies | Dependency graph validation (DAG check) | `twin-orchestrator` |\n| Priority inversion | `critical` intents bypass normal queue | `hermes` (queue priority) |\n| Unauthorized arbitration | Only `twin-orchestrator` instances may arbitrate | `govern` + RBAC |\n\n---\n\n## E.5 SDK Variants\n\n### E.5.1 SDK Taxonomy\n\nThe Bridge exposes **five SDK variants** — same underlying contract, different fidelity and bundle size. Each variant is a separate npm package under the `@iw` scope.\n\n```\n┌─────────────────────────────────────────────────────────────────────────────┐\n│ SDK VARIANT SPECTRUM │\n│ │\n│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────┐ ┌──────┐ │\n│ │ @iw/sdk-full│ │@iw/sdk-lite │ │@iw/sdk-content│ │@iw/sdk- │ │@iw/ │ │\n│ │ │ │ │ │ │ │ minimal │ │sdk- │ │\n│ │ ~180 KB │ │ ~85 KB │ │ ~45 KB │ │ ~12 KB │ │head- │ │\n│ │ gzipped │ │ gzipped │ │ gzipped │ │ gzipped │ │less │ │\n│ │ │ │ │ │ │ │ │ │~3 KB │ │\n│ └──────┬──────┘ └──────┬──────┘ └──────┬──────┘ └────┬────┘ └──┬───┘ │\n│ │ │ │ │ │ │\n│ ▼ ▼ ▼ ▼ ▼ │\n│ Full-stack Browser/Node Content-only Telemetry Server- │\n│ SDK with SDK with SDK (read-only + Events side only │\n│ all features streaming Discovery + (no UI) │\n│ + UI helpers + caching Projections) │\n│ │\n│ Use case: Use case: Use case: Use case: Use case: │\n│ Workbench Web apps, Mobile apps, Analytics Background │\n│ builders, dashboards, embeddable pipelines, workers, │\n│ extension CLI tools widgets logging edge │\n│ developers functions│\n│ │\n└─────────────────────────────────────────────────────────────────────────────┘\n```\n\n### E.5.2 SDK Interface Contracts\n\n```typescript\n// ============================================================\n// @iw/sdk-full — The complete SDK (workbench + headless)\n// ============================================================\n\n\nexport interface SDKFullConfig {\n /** Bridge endpoint */\n baseUrl: string; // default: \"https://api.integratewise.ai\"\n\n /** Authentication */\n jwt: string;\n\n /** Tenant context (optional — extracted from JWT if omitted) */\n tenantId?: string;\n\n /** MCP endpoint for AI assistant integration */\n mcpEndpoint?: string; // default: \"https://mcp.integratewise.ai/v1/mcp\"\n\n /** Enable React UI components */\n enableUI: boolean;\n\n /** UI theme config */\n theme?: UIThemeConfig;\n\n /** Real-time transport: 'sse' | 'ws' | 'polling' */\n realtimeTransport: \"sse\" | \"ws\" | \"polling\";\n\n /** Maximum reconnect attempts */\n maxReconnects: number;\n\n /** Debug mode */\n debug: boolean;\n}\n\nexport class IWBridgeFull {\n constructor(config: SDKFullConfig);\n\n // ── Core Capabilities ──\n invoke(capability: string, payload: unknown): Promise>;\n discover(): Promise;\n stream(capability: string, payload: unknown): AsyncIterable;\n\n // ── Projections ──\n getProjection(name: string, params?: Record): Promise;\n listProjections(): Promise;\n\n // ── Memory ──\n queryMemory(query: MemoryQuery): Promise;\n shareKnowledge(packet: KSPPacketInput): Promise;\n\n // ── Agent ──\n getTwin(): TwinSession;\n listAgents(): Promise;\n publishIntent(intent: A2AIntentInput): Promise;\n\n // ── Governance ──\n getProposals(filter?: ProposalFilter): Promise;\n approveProposal(proposalId: string): Promise;\n rejectProposal(proposalId: string, reason?: string): Promise;\n\n // ── Real-time ──\n onEvent(event: string, handler: EventHandler): Unsubscribe;\n onIntent(handler: IntentHandler): Unsubscribe;\n\n // ── UI Components (React) ──\n Workbench: typeof WorkbenchUI;\n TwinPanel: typeof TwinPanel;\n MemoryExplorer: typeof MemoryExplorer;\n ProposalQueue: typeof ProposalQueue;\n}\n\n// ============================================================\n// @iw/sdk-lite — Browser/Node SDK (no UI, with streaming)\n// ============================================================\n\n\nexport interface SDKLiteConfig {\n baseUrl: string;\n jwt: string;\n tenantId?: string;\n realtimeTransport: \"sse\" | \"polling\";\n cacheEnabled: boolean;\n cacheTtlSeconds: number;\n}\n\nexport class IWBridgeLite {\n constructor(config: SDKLiteConfig);\n\n invoke(capability: string, payload: unknown): Promise>;\n stream(capability: string, payload: unknown): AsyncIterable;\n discover(): Promise;\n queryMemory(query: MemoryQuery): Promise;\n onEvent(event: string, handler: EventHandler): Unsubscribe;\n\n // Lite has no UI components, no Twin session, no governance actions\n}\n\n// ============================================================\n// @iw/sdk-content — Content/readonly SDK (smallest interactive)\n// ============================================================\n\n\nexport interface SDKContentConfig {\n baseUrl: string;\n jwt: string;\n tenantId?: string;\n}\n\nexport class IWContentSDK {\n constructor(config: SDKContentConfig);\n\n // Read-only operations only\n discover(): Promise;\n getProjection(name: string): Promise;\n queryMemory(query: MemoryQuery): Promise;\n\n // No invoke, no stream, no write operations\n}\n\n// ============================================================\n// @iw/sdk-minimal — Telemetry + Events only\n// ============================================================\n\n\nexport interface SDKMinimalConfig {\n baseUrl: string;\n jwt: string;\n tenantId?: string;\n}\n\nexport class IWMinimalSDK {\n constructor(config: SDKMinimalConfig);\n\n // Events and telemetry only\n track(event: string, properties?: Record): Promise;\n onEvent(event: string, handler: EventHandler): Unsubscribe;\n health(): Promise;\n}\n\n// ============================================================\n// @iw/sdk-headless — Server-side / edge functions\n// ============================================================\n\n\nexport interface SDKHeadlessConfig {\n baseUrl: string;\n /** API Key (server-side only) OR JWT */\n apiKey?: string;\n jwt?: string;\n tenantId: string; // Required for headless\n}\n\nexport class IWHeadlessSDK {\n constructor(config: SDKHeadlessConfig);\n\n // Full capability invocation (server-side has no governance gate —\n // assumes calling service has already gated)\n invoke(capability: string, payload: unknown): Promise>;\n\n // Batch operations\n invokeBatch(requests: CapabilityInvokeRequest[]): Promise;\n\n // Admin operations (requires admin scope)\n getTenantConfig(): Promise;\n updateTenantConfig(config: Partial): Promise;\n\n // No real-time, no UI, no streaming (server-side is request/response)\n}\n```\n\n### E.5.3 SDK Package Structure\n\n```\npackages/\n├── sdk-full/\n│ ├── src/\n│ │ ├── index.ts # Main export: IWBridgeFull\n│ │ ├── client.ts # HTTP client + retry logic\n│ │ ├── streaming.ts # SSE/WebSocket stream handler\n│ │ ├── capabilities.ts # Capability invocation wrappers\n│ │ ├── projections.ts # Projection fetch + cache\n│ │ ├── memory.ts # Memory query + KSP\n│ │ ├── agent.ts # Twin session + A2A intents\n│ │ ├── governance.ts # Proposal review actions\n│ │ ├── realtime.ts # Event bus (SSE/WS/polling)\n│ │ ├── ui/ # React components\n│ │ │ ├── Workbench.tsx\n│ │ │ ├── TwinPanel.tsx\n│ │ │ ├── MemoryExplorer.tsx\n│ │ │ └── ProposalQueue.tsx\n│ │ └── types.ts # Re-exports from @iw/sdk-types\n│ ├── package.json\n│ └── tsconfig.json\n│\n├── sdk-lite/\n│ ├── src/\n│ │ ├── index.ts\n│ │ ├── client.ts # Shared HTTP logic (same as sdk-full)\n│ │ ├── streaming.ts # SSE only\n│ │ ├── capabilities.ts\n│ │ ├── projections.ts\n│ │ ├── memory.ts\n│ │ └── realtime.ts\n│ └── package.json\n│\n├── sdk-content/\n│ ├── src/\n│ │ ├── index.ts\n│ │ ├── client.ts # Simple fetch wrapper\n│ │ ├── discovery.ts\n│ │ ├── projections.ts\n│ │ └── memory.ts # Read-only queries\n│ └── package.json\n│\n├── sdk-minimal/\n│ ├── src/\n│ │ ├── index.ts\n│ │ ├── client.ts # Tiny fetch wrapper (~2KB)\n│ │ ├── events.ts\n│ │ └── health.ts\n│ └── package.json\n│\n├── sdk-headless/\n│ ├── src/\n│ │ ├── index.ts\n│ │ ├── client.ts # Node.js fetch / undici\n│ │ ├── capabilities.ts # Full invoke + batch\n│ │ └── admin.ts # Tenant management\n│ └── package.json\n│\n└── sdk-types/ # SHARED types package (all SDKs depend on this)\n ├── src/\n │ ├── capability.ts\n │ ├── discovery.ts\n │ ├── ksp.ts\n │ ├── a2a.ts\n │ ├── memory.ts\n │ ├── governance.ts\n │ └── events.ts\n └── package.json\n```\n\n### E.5.4 SDK Authentication Patterns\n\n```typescript\n// packages/sdk-types/src/auth.ts\n\n/**\n * Authentication contract for all SDK variants.\n * The Gateway is the single source of truth for auth.\n */\nexport interface SDKAuth {\n /** JWT issued by Gateway (primary auth for browser clients) */\n jwt?: string;\n\n /** API Key for server-to-server calls */\n apiKey?: string;\n\n /** Tenant ID (extracted from JWT if omitted) */\n tenantId?: string;\n\n /** User ID (extracted from JWT if omitted) */\n userId?: string;\n}\n\n/**\n * Every SDK request carries these headers:\n *\n * Authorization: Bearer \n * x-tenant-id: ← injected by Gateway, never trusted from client\n * x-correlation-id: ← for distributed tracing\n * x-sdk-variant: \n * x-sdk-version: \n */\n```\n\n---\n\n## E.6 Ecosystem App Composition Patterns\n\n### E.6.1 The Composition Model\n\n> **Third-party apps compose on the Bridge by consuming capabilities, not by importing internal services.** The Bridge is a capability platform, not a PaaS.\n\n```\n┌─────────────────────────────────────────────────────────────────────────────┐\n│ ECOSYSTEM APP COMPOSITION PATTERNS │\n│ │\n│ ┌─────────────────────────────────────────────────────────────────────┐ │\n│ │ THIRD-PARTY APP (e.g., CS Widget) │ │\n│ │ │ │\n│ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │ │\n│ │ │ @iw/sdk │ │ @iw/sdk │ │ @iw/sdk │ │ │\n│ │ │ -content │ │ -lite │ │ -full │ │ │\n│ │ │ (read-only)│ │ (streaming)│ │ (workbench)│ │ │\n│ │ └──────┬──────┘ └──────┬──────┘ └──────┬──────┘ │ │\n│ │ │ │ │ │ │\n│ │ └─────────────────┴─────────────────┘ │ │\n│ │ │ │ │\n│ │ ┌──────┴──────┐ │ │\n│ │ │ Gateway │ │ │\n│ │ │ (JWT gate) │ │ │\n│ │ └──────┬──────┘ │ │\n│ │ │ │ │\n│ └───────────────────────────┼─────────────────────────────────────────┘ │\n│ │ │\n│ ┌───────────────────────────┼─────────────────────────────────────────┐ │\n│ │ CONTINUITY BRIDGE │ │ │\n│ │ ▼ │ │\n│ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │ │\n│ │ │ Discovery │ │ Capability │ │ Events │ │ │\n│ │ │ (GET /api/ │ │ (POST /api/│ │ (SSE /api/ │ │ │\n│ │ │ v1/disco) │ │ v1/cap/...)│ │ v1/stream) │ │ │\n│ │ └─────────────┘ └─────────────┘ └─────────────┘ │ │\n│ │ │ │\n│ │ The app NEVER imports: services/, packages/spine-schema/, D1 │ │\n│ │ The app NEVER calls: act, pipeline, normalizer directly │ │\n│ │ │ │\n│ └─────────────────────────────────────────────────────────────────────┘ │\n│ │\n│ COMPOSITION PATTERNS: │\n│ 1. Embed Pattern — iframe/widget using @iw/sdk-content │\n│ 2. Plugin Pattern — browser extension using @iw/sdk-lite │\n│ 3. Workbench Pattern — full app using @iw/sdk-full │\n│ 4. Webhook Pattern — server receives events, uses @iw/sdk-headless │\n│ 5. MCP Pattern — AI assistant connects via MCP, no SDK needed │\n│ │\n└─────────────────────────────────────────────────────────────────────────────┘\n```\n\n### E.6.2 Composition Patterns Catalog\n\n#### Pattern 1: Embed Pattern (iframe/widget)\n\n```typescript\n// Third-party app embeds a Bridge-powered widget\n\nconst sdk = new IWContentSDK({\n baseUrl: \"https://api.integratewise.ai\",\n jwt: getTenantJWT(), // From host app auth\n});\n\n// Fetch discovery to learn what's available\nconst discovery = await sdk.discover();\n\n// Render a read-only account health widget\nconst projection = await sdk.getProjection(\"cs-health\", {\n accountId: props.accountId,\n});\n\n// Mount React component (if using @iw/sdk-full)\n// Or render raw HTML from projection data\n```\n\n**Security:** Embed runs in sandboxed iframe. Host app provides JWT via `postMessage`. SDK validates JWT on every call.\n\n#### Pattern 2: Plugin Pattern (browser extension)\n\n```typescript\n// Browser extension content script\n\nconst bridge = new IWBridgeLite({\n baseUrl: \"https://api.integratewise.ai\",\n jwt: await chrome.storage.local.get(\"iw_jwt\"),\n realtimeTransport: \"sse\",\n cacheEnabled: true,\n});\n\n// Listen for Twin proposals while user browses Salesforce\nbridge.onEvent(\"twin.proposal\", (proposal) => {\n showInPageNotification(proposal);\n});\n\n// Invoke capability from any page\nbridge.invoke(\"salesforce.opportunity.update\", {\n opportunityId: extractIdFromPage(),\n stage: \"Closed Won\",\n});\n```\n\n**Security:** Extension manifest declares `api.integratewise.ai` as allowed host. CORS preflight validated by Gateway.\n\n#### Pattern 3: Workbench Pattern (full app)\n\n```typescript\n// Custom workbench built on Bridge\n\nconst bridge = new IWBridgeFull({\n baseUrl: 'https://api.integratewise.ai',\n jwt: auth.jwt,\n enableUI: true,\n theme: customTheme,\n realtimeTransport: 'ws',\n});\n\n// Render full workbench with Twin panel\nfunction App() {\n return (\n
\n \n \n \n \n
\n );\n}\n```\n\n**Security:** Full SDK requires `scope: workbench` in JWT. UI components are sandboxed React trees with CSP.\n\n#### Pattern 4: Webhook Pattern (server-side integration)\n\n```typescript\n// Third-party server receives Bridge webhooks\n\nconst bridge = new IWHeadlessSDK({\n baseUrl: \"https://api.integratewise.ai\",\n apiKey: process.env.IW_API_KEY,\n tenantId: process.env.IW_TENANT_ID,\n});\n\n// Webhook handler\napp.post(\"/webhooks/iw\", async (req, res) => {\n const event = req.body;\n\n // Verify webhook signature\n verifyWebhookSignature(req);\n\n switch (event.type) {\n case \"proposal.approved\":\n await bridge.invoke(\"slack.notify\", {\n channel: \"#ops\",\n message: `Proposal approved: ${event.proposalId}`,\n });\n break;\n\n case \"memory.promoted\":\n await syncToExternalKB(event.memory);\n break;\n }\n\n res.sendStatus(200);\n});\n```\n\n**Security:** Webhook signatures use HMAC-SHA256. API keys are server-side only. Never exposed to browser.\n\n#### Pattern 5: MCP Pattern (AI assistant integration)\n\n```json\n// No SDK needed — AI assistant connects via MCP directly\n{\n \"mcpServers\": {\n \"integratewise\": {\n \"type\": \"sse\",\n \"url\": \"https://mcp.integratewise.ai/v1/mcp\",\n \"headers\": {\n \"Authorization\": \"Bearer ${IW_JWT}\"\n }\n }\n }\n}\n```\n\n**Security:** MCP connection is read-only by default. Write capabilities require `scope: write` + governance approval.\n\n### E.6.3 App Manifest Contract\n\n```typescript\n// packages/sdk-types/src/manifest.ts\n\n/**\n * Ecosystem App Manifest — every third-party app registers with the Bridge.\n */\nexport interface EcosystemAppManifest {\n /** Unique app ID (reverse DNS) */\n id: string; // e.g., \"com.acme.cswidget\"\n\n /** Human name */\n name: string;\n\n /** Description for Discovery */\n description: string;\n\n /** App type */\n type: \"embed\" | \"plugin\" | \"workbench\" | \"webhook\" | \"mcp-client\";\n\n /** Required SDK variant */\n sdkVariant: \"content\" | \"lite\" | \"full\" | \"headless\";\n\n /** Minimum SDK version */\n minSdkVersion: string;\n\n /** Required scopes */\n requiredScopes: string[];\n\n /** Capabilities this app consumes */\n consumedCapabilities: string[];\n\n /** Capabilities this app provides (if any) */\n providedCapabilities?: Capability[];\n\n /** Webhook URL (for webhook-type apps) */\n webhookUrl?: string;\n\n /** Content Security Policy */\n csp: {\n allowedOrigins: string[];\n allowedConnectSrc: string[];\n };\n\n /** Data handling */\n dataHandling: {\n storesDataLocally: boolean;\n cachesInBrowser: boolean;\n piiAccess: boolean;\n };\n\n /** Registration */\n registeredBy: string; // User ID\n registeredAt: string;\n status: \"pending\" | \"approved\" | \"rejected\" | \"revoked\";\n}\n```\n\n---\n\n## E.7 Template Starter (`create-iw-app`)\n\n### E.7.1 Template Architecture\n\n```\ncreate-iw-app/\n├── templates/\n│ ├── embed-widget/ # Pattern 1: iframe/widget\n│ │ ├── package.json\n│ │ ├── src/\n│ │ │ ├── main.ts # Widget entry point\n│ │ │ ├── bridge.ts # @iw/sdk-content init\n│ │ │ └── widget.tsx # React widget component\n│ │ └── README.md\n│ │\n│ ├── browser-extension/ # Pattern 2: browser plugin\n│ │ ├── manifest.json\n│ │ ├── src/\n│ │ │ ├── background.ts # Service worker\n│ │ │ ├── content.ts # Content script\n│ │ │ ├── bridge.ts # @iw/sdk-lite init\n│ │ │ └── popup.tsx # Extension popup\n│ │ └── README.md\n│ │\n│ ├── workbench-app/ # Pattern 3: full workbench\n│ │ ├── package.json\n│ │ ├── src/\n│ │ │ ├── App.tsx\n│ │ │ ├── bridge.ts # @iw/sdk-full init\n│ │ │ ├── pages/\n│ │ │ │ ├── Dashboard.tsx\n│ │ │ │ ├── TwinPage.tsx\n│ │ │ │ └── Settings.tsx\n│ │ │ └── components/\n│ │ └── README.md\n│ │\n│ ├── webhook-server/ # Pattern 4: server-side webhook\n│ │ ├── package.json\n│ │ ├── src/\n│ │ │ ├── server.ts # Express/Fastify server\n│ │ │ ├── bridge.ts # @iw/sdk-headless init\n│ │ │ ├── webhooks.ts # Webhook handlers\n│ │ │ └── verify.ts # Signature verification\n│ │ └── README.md\n│ │\n│ └── mcp-client/ # Pattern 5: AI assistant client\n│ ├── package.json\n│ ├── src/\n│ │ ├── client.ts # MCP client setup\n│ │ └── prompts/ # System prompts for Bridge\n│ └── README.md\n│\n├── src/\n│ ├── index.ts # CLI entry: create-iw-app