kai-email-system
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The discovery phase instructs the agent to read potentially sensitive areas of the codebase to build context.
- Evidence: The instructions direct the agent to "Check for user flows, auth systems, billing integrations, database schemas, API routes" during the Phase 0 discovery process. This may lead to the exposure of sensitive technical architecture or metadata within the generated MARKETING.md file.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project data and external marketing copy to generate its product bible and emails, creating a vulnerability surface.
- Ingestion points: Reads local project files such as README.md, CLAUDE.md, and package.json, as well as external landing pages and marketing materials.
- Boundary markers: Absent; the skill lacks specific delimiters or instructions to ignore embedded commands within the ingested data sources.
- Capability inventory: The skill has the capability to read local files and write generated email systems to the workspace/ and project root directories.
- Sanitization: Absent; ingested content is processed and used directly to populate the product bible and subsequent email templates without validation or filtering.
Audit Metadata