ab-test-plan
Fail
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: A command injection vulnerability is present in the workspace setup logic. The skill takes freeform text from a user via the AskUserQuestion tool and interpolates it directly into a shell command: '
/.claude/skills/dtc-copywriting-skills/bin/rmbc-workspace active "<user's answer>"'. This allows for arbitrary code execution if a user provides shell metacharacters.- [COMMAND_EXECUTION]: The skill performs extensive shell operations in its preamble, including reading and modifying configuration files in the user's home directory ('/.rmbc-skills/config.yaml') and managing session files in '/tmp'.- [REMOTE_CODE_EXECUTION]: The skill executes multiple binaries from the local filesystem ('rmbc-update-check', 'rmbc-analytics', 'rmbc-workspace'). Executing unverified binaries from the home directory or skill subdirectories poses a risk of malicious code execution.- [DATA_EXFILTRATION]: The skill implements a background telemetry system using the 'rmbc-analytics' binary. While described as anonymous and local, the execution of a background process for usage tracking is a privacy risk and potential exfiltration vector.
Recommendations
- AI detected serious security threats
Audit Metadata