05-copy-quang-cao
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from local context files (
.agents/product-marketing-context.mdand.agents/personal-brand-context.md) to determine the operation mode and tailor the generated advertisement copy. If these files are populated with malicious instructions (e.g., via a compromised repository), the agent may be manipulated into generating deceptive content or including malicious links. - Ingestion points:
.agents/product-marketing-context.mdand.agents/personal-brand-context.mdinSKILL.md. - Boundary markers: Absent; the skill reads the files to determine the mode without specific delimiters for user-provided content.
- Capability inventory: Generates complex marketing copy, headlines, and descriptions which could be influenced by injected instructions.
- Sanitization: No sanitization or validation of the context file contents was observed.
Audit Metadata