drush
Warn
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill documents the use of
drush php:evalanddrush php:script. These commands allow the execution of arbitrary PHP code within the Drupal bootstrap context, granting the agent full access to the Drupal API and underlying system resources. - [COMMAND_EXECUTION]: The skill provides instructions for powerful administrative commands such as
drush sql:queryfor direct database manipulation,drush user:passwordfor account resets, anddrush sql:dropfor deleting database tables. - [DATA_EXFILTRATION]: Instructions for
drush sql:dumpanddrush config:getprovide a mechanism to extract the entire site database or sensitive configuration values from the environment. - [INDIRECT_PROMPT_INJECTION]: The skill defines a large attack surface for indirect prompt injection. If an agent uses these commands to process untrusted data (e.g., executing PHP logic derived from a user request or external content), it could be manipulated into performing unauthorized actions.
- Ingestion points: User prompts requesting Drupal administrative tasks or site troubleshooting.
- Boundary markers: None identified; there are no instructions to sanitize or wrap user inputs before they are passed to the Drush CLI.
- Capability inventory: Full PHP execution (
php:eval), full SQL access (sql:query), user management (user:login,user:password), and database destruction (sql:drop). - Sanitization: No sanitization or input validation logic is present in the skill instructions.
Audit Metadata