shell-scripting

Warn

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill documentation includes an example command that accesses the sensitive system file /etc/passwd. While intended to demonstrate text processing with specific delimiters, referencing sensitive system files is a risky practice as it normalizes accessing sensitive host data in shell commands.\n
  • Evidence: awk -F: '{print $1}' /etc/passwd # Custom delimiter in SKILL.md.\n- [INDIRECT_PROMPT_INJECTION]: The skill provides various patterns for ingesting and processing external data from sources like logs, CSV, and JSON files. This creates a surface for indirect prompt injection if an agent processes attacker-controlled data using these instructions without proper sanitization.\n
  • Ingestion points: File reading loops and text processing patterns in SKILL.md (e.g., while read -r line; do ... done < "$file", jq '.key' file.json).\n
  • Boundary markers: The examples do not include boundary markers or instructions to ignore embedded commands within the data.\n
  • Capability inventory: The skill documents the use of powerful shell utilities including find, grep, jq, awk, sed, mkdir, and rm.\n
  • Sanitization: There is no evidence of input validation or sanitization in the provided file processing examples.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 6, 2026, 10:03 AM
Security Audit — agent-trust-hub — shell-scripting