product-behavioral-nudge-engine
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it ingests and processes untrusted data from user task queues and profiles.
- Ingestion points: Untrusted data enters the agent context through the analysis of user task queues (
pendingTasks) and behavioral profiles (userProfile), as seen in the 'Workflow Process' (Phase 2) and the TypeScript example inSKILL.md. - Boundary markers: There are no explicit boundary markers or delimiters defined in the instructions to prevent the agent from following instructions embedded within the user's tasks.
- Capability inventory: The skill generates actionable outputs such as SMS/Email drafts and 'Micro-Sprint' prompts which are delivered to the user; if an ingested task contains a malicious prompt, the agent might inadvertently include it in these outgoing communications.
- Sanitization: There is no mention of sanitization, filtering, or validation logic to ensure that task content does not override the behavioral engine's instructions.
Audit Metadata