skills/vercel-labs/bash-tool/csv/Gen Agent Trust Hub

csv

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • Bash Tool Integration: The skill utilizes standard Unix utilities such as awk, cut, and sort to perform local data transformations. These tools are used within their intended scope for text and record processing.
  • Indirect Prompt Injection Surface: The skill processes data from external CSV files, which presents a surface for potential indirect prompt injection where content in a file might attempt to influence an AI agent.
  • Ingestion points: Data is ingested through the $FILE argument in analyze.sh, filter.sh, select.sh, and sort.sh.
  • Boundary markers: The scripts do not implement specific delimiters or 'ignore' instructions to isolate CSV data from the agent's context.
  • Capability inventory: The skill's capabilities are limited to reading and processing files; it lacks network access or file-writing permissions.
  • Sanitization: The implementation uses shell variable quoting and awk's -v variable passing to handle inputs, which helps mitigate basic shell injection risks.
  • Secure Scripting Practices: The scripts are written with standard security considerations for shell environments. For instance, file paths are quoted to handle special characters correctly, and command-line flags in the sorting script are explicitly validated through a case statement.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 02:55 PM
Security Audit — agent-trust-hub — csv