The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill involves ingesting and processing data from external, potentially untrusted sources, which defines an indirect prompt injection attack surface.
Ingestion points: Instructions in SKILL.md direct the agent to import data from local CSV files (modules/custom/my_module/data/products.csv) and remote JSON API endpoints (https://api.example.com/users).
Boundary markers: The provided YAML migration templates do not include delimiters or specific instructions for the AI to disregard commands or instructions that may be embedded within the source data.
Capability inventory: The skill executes drush and composer commands and allows for the creation of custom PHP plugins, providing a highly capable execution environment for data being processed.
Sanitization: There is no evidence of explicit sanitization, filtering, or validation of the ingested content in the provided PHP or YAML templates before it is transformed and stored in the destination system.

drupal-migration