Workflow Automation Platform Design

Core Functional Scope

• DAG-Based Execution: Workflows are modeled as Directed Acyclic Graphs. The engine resolves dependencies via topological sort, enabling parallel branch execution and deterministic join semantics.
• Node Plugin Architecture: Every integration (Slack, Gmail, HTTP, DB) is a self-contained plugin implementing a standard NodeInterface. The Node Registry catalogs all available plugins with versioning and schema metadata.
• Credential Management: Per-user encrypted credential storage using AES-256. OAuth2 tokens are automatically refreshed before expiry. Credentials are injected at execution time, never stored in workflow definitions.
• Multi-Trigger System: Workflows can be initiated via webhooks (HTTP POST), cron schedules, pub/sub event listeners, or manual API calls. Each trigger type implements a standard TriggerInterface.
• Expression Evaluation: Data transformation between nodes uses JSONata expressions or sandboxed JavaScript. The Transform Engine evaluates expressions in an isolated V8 context to prevent side-effects.

Standardized Node Interface & Results

interface INode {
    execute(
        input: INodeExecutionData[],
        context: IExecutionContext
    ): Promise<INodeExecutionData[]>;
}

enum ExecutionStatus {
    SUCCESS,
    RETRYABLE_ERROR,   // Rate limit, 503
    PERMANENT_ERROR,   // Bad schema, 400
    WAITING,           // Webhook callback
    CANCELLED          // User abort
}