The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by defining a workflow where data from one agent influences the execution of another with high capabilities.
Ingestion points: The workflow consumes plans produced by preceding skills (drupal-brainstorming, writing-plans) as the primary input for sub-agent prompts in SKILL.md.
Boundary markers: There are no instructions in SKILL.md to use delimiters or specific security markers to separate untrusted plan data from system instructions in the delegated prompts.
Capability inventory: According to the delegation rules in SKILL.md, sub-agents invoked (such as @drupal-builder and @drupal-verifier) have the capability to modify the filesystem and execute system commands (e.g., drush eval).
Sanitization: The skill lacks explicit requirements for sanitizing or validating plan content before it is processed by the implementation agents.

drupal-delegation