csv-data-analyst
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONNO_CODE
Full Analysis
- [PROMPT_INJECTION] (HIGH): Vulnerable to Indirect Prompt Injection. The skill is designed to ingest and summarize untrusted external CSV data. It lacks explicit boundary markers or instructions to ignore embedded commands within the data. Evidence: 1. Ingestion points: user-referenced CSV files via
file_path. 2. Boundary markers: Absent. 3. Capability inventory: Python execution (pandas/matplotlib). 4. Sanitization: Absent.- [NO_CODE] (HIGH): The filescripts/analyze.pywhich contains the core logic for thesummarize_csvfunction is missing from the provided skill package. This prevents verification of whether the script performs dangerous operations like unauthorized network access or file modifications.- [COMMAND_EXECUTION] (MEDIUM): The skill utilizes a Python environment to process data. While intended for analysis, the execution of logic on user-provided files constitutes a significant attack surface if the underlying script lacks rigorous input validation.- [PROMPT_INJECTION] (LOW): The markdown contains aggressive behavioral overrides in the 'CRITICAL BEHAVIOR REQUIREMENT' section, instructing the agent to bypass standard conversational protocols and act without user confirmation. While functional for the skill, these patterns mirror high-pressure prompt injection techniques.
Recommendations
- AI detected serious security threats
Audit Metadata