recurring-loops
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external marketing inputs and triggers, which creates a surface for indirect prompt injection where malicious instructions could be embedded in the data.
- Ingestion points: Accesses external data through user-defined
inputs,source_freshnessmarkers, and the shared.agents/marketing-context.mdfile. - Boundary markers: Implements significant mitigation strategies including mandatory "dry run" fixed-snapshots, hierarchical "approval layers" (Draft, Approve, Mutate), and explicit instructions to "fail closed" when authority is unclear.
- Capability inventory: Conceptually designed to "mutate external state", "send", "publish", and "spend" via agent tools, which could be misused if injected instructions are followed.
- Sanitization: Relies on idempotency keys, verification/reconciliation phases, and monitoring to ensure actions match intended state transitions.
Audit Metadata