attribution-model
Pass
Audited by Gen Agent Trust Hub on Feb 27, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect prompt injection surface via local data ingestion.
- Ingestion points: The skill reads multiple files from
~/.claude-marketing/brands/, including brand profiles, manifests, and compliance guidelines, as described in the Process section. - Boundary markers: No delimiters or explicit instructions are provided to ensure the agent ignores any malicious prompts that might be embedded within the loaded JSON or Markdown files.
- Capability inventory: The agent's capabilities are confined to generating text-based marketing strategies and implementation guides; it does not have access to network operations or system command execution.
- Sanitization: There is no indication that the ingested brand data is sanitized or validated against a strict schema before being added to the model context.
Audit Metadata