constructive-realtime

Pass

Audited by Gen Agent Trust Hub on Aug 27, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill describes standard, secure patterns for implementing real-time updates via database triggers and GraphQL subscriptions. It provides clear documentation for configuring the Constructive platform's built-in realtime capabilities without introducing any external dependencies or dangerous command executions. \n- [INDIRECT_PROMPT_INJECTION]: The skill defines a system that ingests and broadcasts database changes, which constitutes an indirect data flow surface. \n
  • Ingestion points: DML operations (INSERT, UPDATE, DELETE) on database tables configured with the DataRealtime node, as documented in references/realtime-subscriptions.md. \n
  • Boundary markers: The architecture uses derived Row-Level Security (RLS) to ensure that subscribers only receive data they are authorized to access. It also includes overflow markers to handle large payloads safely. \n
  • Capability inventory: The backend provisioner automates the creation of a subscriptions_public schema, partitioned change_log tables, and statement-level triggers for event emission. \n
  • Sanitization: Subscription security is strictly governed by inheriting the source table's SELECT policies, effectively sanitizing the delivery channel based on established access control rules.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 27, 2026, 01:07 PM
Security Audit — agent-trust-hub — constructive-realtime