The Agent Skills Directory

[PROMPT_INJECTION]: The skill directs the agent to research external modules and issue queues on drupal.org to determine project requirements.
Ingestion points: The "Research-First Philosophy" section in SKILL.md instructs the agent to browse and evaluate content from drupal.org.
Boundary markers: The skill does not provide specific delimiters or instructions to treat external research data as untrusted.
Capability inventory: The agent is granted capabilities to write files and execute shell commands (via Drush and PHPUnit).
Sanitization: There are no instructions to validate or sanitize information gathered from external sources before it is used to generate project code or configuration.
[COMMAND_EXECUTION]: The skill provides extensive documentation for using the Drush CLI tool to scaffold modules, entities, and configurations.
Instructions in SKILL.md suggest using drush generate and drush field:create to automate the creation of PHP files and configuration YAMLs.
It provides specific commands to run Drush non-interactively using --answers with JSON and execute PHP code directly via drush php:eval.
While these are standard developer workflows for Drupal, the generation and execution of code based on LLM interpretation of these patterns constitutes dynamic execution and requires developer oversight.

drupal-expert