NYC

data-analysis

Fail

Audited by Gen Agent Trust Hub on Feb 15, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • Indirect Prompt Injection (HIGH): The skill possesses a high-risk vulnerability surface by combining external data ingestion with powerful execution capabilities.\n
  • Ingestion points: Untrusted data is ingested from data.csv using pd.read_csv and from SQL databases through query templates (SKILL.md, Steps 1 and 2).\n
  • Boundary markers: Absent. The skill provides no instructions or delimiters to help the agent distinguish between legitimate data and malicious instructions embedded within the processed datasets.\n
  • Capability inventory: The skill allows high-privilege tools including Bash (allowed-tools) and Python execution, which includes file-writing operations via plt.savefig (SKILL.md, Step 4).\n
  • Sanitization: Absent. There is no logic or guidance for validating, escaping, or sanitizing external content before it is processed by code or used to influence agent decisions.\n- COMMAND_EXECUTION (MEDIUM): The skill explicitly allows the Bash tool in its configuration. While common for data tasks, the inclusion of arbitrary shell access significantly escalates the impact of potential prompt injection vulnerabilities found in the analysis scripts.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 15, 2026, 09:19 PM