mckinsey-consultant
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill utilizes external data obtained through
web_search(Phase 1, Step 3 and Phase 3, Step 6) to populate business analysis frameworks and generate PowerPoint reports. This creates a vulnerability surface where instructions embedded in third-party websites could be ingested by the agent and influence its reasoning or output.\n - Ingestion points: Untrusted external data enters the agent's context during the Hypothesis formation and Data Collection phases described in
SKILL.md.\n - Boundary markers: There are no specific instructions in the skill templates to wrap search results in boundary delimiters (e.g., XML tags) or to explicitly instruct the agent to ignore instructions found within the data.\n
- Capability inventory: The skill possesses extensive capabilities including file generation (PowerPoint and Excel), local file reading, and performing recursive network-based searches, which are high-value targets for multi-step injection attacks.\n
- Sanitization: While the methodology emphasizes "multi-source verification" for factual accuracy, it does not provide security-focused sanitization instructions to prevent the agent from executing commands hidden in web content.
Audit Metadata