ad-creative-audit

Warn

Audited by Gen Agent Trust Hub on Aug 11, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes several internal shell scripts (rmbc-update-check, rmbc-analytics, rmbc-workspace) during its preamble and post-execution phases. These scripts are invoked from paths relative to the skill installation or the user's home directory.
  • [COMMAND_EXECUTION]: The skill contains a potential command injection vulnerability in the workspace activation logic. It instructs the agent to execute /bin/rmbc-workspace active "<user's answer>" using freeform text input from the user. An attacker could provide input designed to break out of the quotes and execute arbitrary shell commands.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests and processes untrusted data (ad_copy, target_audience, performance_data) without implementing boundary markers or sanitization instructions.
  • Ingestion points: Inputs ad_copy, target_audience, visual_description, and performance_data in SKILL.md.
  • Boundary markers: None identified; untrusted content is directly scored against framework principles.
  • Capability inventory: The skill has access to shell execution (bash).
  • Sanitization: No escaping or validation of external content is specified before processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 11, 2026, 01:36 PM
Security Audit — agent-trust-hub — ad-creative-audit