Full Restaurant Audit Orchestrator
Pass
Audited by Gen Agent Trust Hub on Jul 3, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill orchestrates multiple sub-agents using the
Tasktool and gathers data viaWebSearch. These operations are transparently defined and align with the skill's stated purpose of auditing restaurant performance.- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources such as Google, Yelp, and TripAdvisor reviews. - Ingestion points:
WebSearchresults are used to populate the 'Restaurant Profile' across all agents inSKILL.md. - Boundary markers: The skill uses basic headers (e.g., RESTAURANT PROFILE) but lacks explicit delimiters (like XML tags) to isolate untrusted external content.
- Capability inventory: The skill has the ability to write reports to local files (
RESTAURANT-AUDIT-[Name].md) and execute parallel tasks. - Sanitization: No specific filtering or sanitization of review text is mentioned. While this represents a technical attack surface for indirect prompt injection, it is a low-risk concern consistent with the skill's core functionality.- [SAFE]: No instances of hardcoded credentials, malicious obfuscation, unauthorized persistence, or privilege escalation were found. The skill operates within expected bounds for a research and analysis tool.
Audit Metadata