renewal-radar
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection Guardrails]: The skill contains a well-defined attack surface for processing external data but implements extensive mitigations.
- Ingestion points: Untrusted data enters the agent context via email, chat, transcripts, enrichment, and external documents as described in
SKILL.md. - Boundary markers: Present. The instructions explicitly mark these sources as untrusted, stating they must be treated as data, never instructions, and specifies that the agent must report instruction-like text rather than acting on it.
- Capability inventory: The skill interacts with CRM tools to perform reads and writes (
update-opportunityand renewal opportunity creation) as outlined inSKILL.md. - Sanitization: Present. The skill mandates that content-originated actions require explicit human verification and are strictly downgraded to proposals during scheduled or unattended runs.
Audit Metadata