customer-journey

Pass

Audited by Gen Agent Trust Hub on Sep 25, 2026

Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill instructions mandate the suppression of warnings and caveats in the final output ("Do not return a draft with a warning on it"). This directive conceals the agent's reliance on assumptions or incomplete information from the user, potentially causing them to rely on unverified logic.
  • [INDIRECT_PROMPT_INJECTION]: The skill is instructed to autonomously "fetch the site or page" provided by the user for research, creating an ingestion point for untrusted external data.
  • Ingestion points: External company websites, user-provided URLs, and exported customer lifecycle transition data.
  • Boundary markers: None. The skill lacks instructions to use delimiters or to disregard instructions embedded in the retrieved research content.
  • Capability inventory: The skill writes to local files (.agents/product-context.md), performs network fetches for research, and configures webhook nodes with arbitrary payloads.
  • Sanitization: No validation or sanitization steps are specified for the data retrieved from external sources.
  • [DATA_EXFILTRATION]: The skill's reference architecture defines a webhook node type capable of making external HTTP requests with configurable headers and bodies. This represents a potential data exfiltration vector if a prompt injection attack influences the agent to embed sensitive customer data or internal context into a webhook payload.
  • [EXTERNAL_DOWNLOADS]: The skill fetches positioning and brand voice information from company websites and public competitive sources to populate the product context.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 25, 2026, 04:06 PM