The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of external and potentially untrusted data through various component props (e.g., data in Grid and tasks in Gantt). This exposes an attack surface where an agent could be influenced by instructions embedded within the processed data. \n- Ingestion points: Data-driven props across components described in files such as grid/index.md, gantt/index.md, and filemanager/index.md. \n- Boundary markers: No explicit delimitation or 'ignore embedded instructions' markers are defined in the component logic. \n- Capability inventory: The skill includes RestDataProvider for network operations (GET, POST, PUT, DELETE) and uses window.open for file handling (download/open). \n- Sanitization: The documentation in filter/FilterQuery.md explicitly warns that getQueryHtml returns raw HTML and recommends using it only in trusted contexts due to the use of dangerouslySetInnerHTML in the accompanying recipe.

svar-react