product-messaging-playbook
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes information from local files and external tool connectors to generate formatted documents.
- Ingestion points: The skill reads data from
/foundation/brain.md, tool connectors (~~call recordings,~~CRM,~~support), and user-pasted text. - Boundary markers: The skill uses explicit tags such as
[MISSING]and[TARGET — not yet measured]to denote unverified or absent information. It also specifies that facts pulled from tools are treated as drafts until confirmed by the user. - Capability inventory: The skill reads local files and appends session logs to
/context/skill-sessions.md. - Sanitization: The instructions do not define specific filtering or sanitization logic to detect or strip instructions that might be embedded in the ingested content.
Audit Metadata