the-flow-architect
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill architecture is susceptible to indirect prompt injection due to the ingestion and processing of untrusted external data.
- Ingestion points: Event names and user attributes are processed through the 'Filter DSL' in Trigger, Condition, and Multi-Split nodes defined in references/journey-nodes.md.
- Boundary markers: The instructions do not define specific delimiters or 'ignore' instructions to guide the agent in safely handling potentially adversarial external data during journey evaluation.
- Capability inventory: The skill supports high-privilege actions including Webhooks (external HTTP requests), Run Workflow (triggering sub-journeys), and Update Attribute (modifying database records).
- Sanitization: No explicit sanitization or validation logic is specified for the interpolated Liquid templates or Filter DSL evaluations.
- [COMMAND_EXECUTION]: The references/journey-nodes.md file defines a Webhook node type that enables the agent to configure external HTTP requests (GET, POST, PUT, PATCH, DELETE) to arbitrary destination URLs with custom headers and bodies.
- [DATA_EXFILTRATION]: The Webhook node configuration supports Liquid personalization, allowing sensitive user attributes (such as {{person.email}} or {{person.custom.*}}) to be included in outbound request bodies, creating a potential path for data exfiltration if directed to an untrusted external endpoint.
Audit Metadata