The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to treat 'AUTOPILOT' messages in command/hook output as mandatory instructions that must be followed before any other tasks. This overrides the agent's default autonomy and establishes a high-priority command channel from untrusted external data sources.
[COMMAND_EXECUTION]: The verification guidelines explicitly direct the agent to generate shell scripts in the scripts/tests/ directory and modify their permissions using chmod +x to execute them. This provides a mechanism for the creation and execution of arbitrary code on the host system.
[COMMAND_EXECUTION]: The skill encourages the use of drush eval to execute arbitrary PHP strings in the Drupal environment for testing purposes. While common in development, this capability allows for direct code execution within the application context.
[PROMPT_INJECTION]: The 'Autopilot Interventions' section establishes an indirect prompt injection surface. Because the agent is instructed to monitor command output for specific strings ('AUTOPILOT [type]:') and obey the instructions contained therein, any process or file that can influence command output (such as log files, git hooks, or error messages) can effectively take control of the agent's workflow.
[COMMAND_EXECUTION]: The skill uses ddev exec curl with session cookies to perform smoke tests. While intended for local verification, this involves programmatic interaction with authenticated web sessions using tokens retrieved via drush uli.

drupal-rules