The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill utilizes ddev drush ws to monitor application logs. These logs typically ingest data from untrusted sources (e.g., error messages triggered by web traffic or malformed requests), which creates a surface where malicious instructions could be introduced into the agent's context.
Ingestion points: Monitoring logs via ddev drush ws in SKILL.md.
Boundary markers: None present to distinguish log data from agent instructions.
Capability inventory: Access to ddev drush php:eval (code execution), composer require (package installation), and ddev xdebug (runtime control).
Sanitization: No sanitization or filtering of log content is specified.
[DYNAMIC_EXECUTION]: The skill provides multiple examples of using ddev drush php:eval to execute arbitrary PHP code strings. While intended for inspection of services and configurations, this tool allows for unconstrained code execution within the Drupal bootstrap context.
[COMMAND_EXECUTION]: The skill facilitates the execution of various shell commands including package management through composer, service inspection via drush, and local network requests using curl to verify cache headers.

drupal-debugging