rfm-customer-segmentation
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- Indirect Prompt Injection (HIGH): The skill ingests untrusted CSV data and processes it using high-privilege tools (Bash, Write). Malicious content in the CSV could influence agent behavior or lead to code execution via the agent's logic. Evidence: 1. Ingestion points:
SKILL.mdinstructions for loading transaction CSV data. 2. Boundary markers: Absent; no delimiters are defined. 3. Capability inventory:Bash,Write,Read, andGlobtools are allowed. 4. Sanitization: Absent; no validation or escaping of data fields is mentioned. - Unverifiable Dependencies & Remote Code Execution (MEDIUM): The skill references missing scripts (
core_analysis.py,visualization.py,report_generator.py) and uses unpinned version ranges (>=) for packages outside the trusted scope, posing a supply chain and verification risk. - Dynamic Execution (MEDIUM): The
examples/basic_usage.mdfile promotes executing dynamically constructed Python code strings viapython -c, which is a high-risk pattern when combined with external data ingestion.
Recommendations
- AI detected serious security threats
Audit Metadata