call-claude

Fail

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION] (HIGH): The skill is vulnerable to shell command injection via the {task_description} variable. Since the input is interpolated directly into a bash heredoc without sanitization, an attacker can provide a payload containing the EOT delimiter to terminate the heredoc and execute arbitrary shell commands on the host system.\n- [COMMAND_EXECUTION] (HIGH): The skill explicitly uses the --dangerously-skip-permissions flag with the claude CLI tool. This flag is designed to bypass built-in security safeguards and permission prompts, allowing the tool to perform potentially harmful or unauthorized actions without user oversight.\n- [PROMPT_INJECTION] (LOW): The skill enables indirect prompt injection by passing unvalidated external content to a command-line interface with elevated permissions.\n
  • Ingestion points: {task_description} variable in SKILL.md.\n
  • Boundary markers: Uses <<EOT heredoc syntax, which is bypassable in this implementation.\n
  • Capability inventory: Execution of system commands and potential for full shell access via injection.\n
  • Sanitization: No input validation or escaping is applied to the task description content.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 17, 2026, 06:07 PM