ask-gemini

Fail

Audited by Gen Agent Trust Hub on Aug 20, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill constructs shell commands by directly interpolating user input into strings, such as gemini -p "{{ARGUMENTS}}" and rcs ask gemini "{{ARGUMENTS}}". This constitutes a command injection vulnerability where a user can break out of the intended command context by including shell metacharacters like semicolons, backticks, or pipes to execute arbitrary system commands.\n- [INDIRECT_PROMPT_INJECTION]: The skill exposes a significant attack surface by ingesting untrusted data and using it to drive powerful local capabilities.\n
  • Ingestion points: User input provided via the {{ARGUMENTS}} placeholder in SKILL.md is the primary entry point.\n
  • Boundary markers: The skill lacks any boundary markers or instructions to the agent to treat the input as literal text rather than executable shell content.\n
  • Capability inventory: The skill utilizes local shell execution via the gemini CLI tool and transitional wrappers like npm run ask:gemini.\n
  • Sanitization: There is no logic present to escape shell characters, perform validation, or otherwise sanitize the input before it is passed to the shell.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Aug 20, 2026, 11:37 AM
Security Audit — agent-trust-hub — ask-gemini