The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The SelectableTable component and associated patterns in SKILL.md and assets/components/selectable-table.tsx represent a vulnerability surface for indirect prompt injection. The UI renders item data (e.g., item.name) directly from the items prop without defining security boundaries or sanitization logic.
Ingestion points: The items array and renderRow prop in SelectableTable allow external data to be displayed in the UI.
Boundary markers: The code templates lack delimiters or instructions to the agent to disregard instructions embedded in the displayed data.
Capability inventory: The skill facilitates data-modifying operations through fetch requests (e.g., in the useBulkDeleteItems hook example).
Sanitization: No data sanitization or validation is demonstrated in the rendering logic.
[EXTERNAL_DOWNLOADS]: The skill documentation lists standard, well-known frontend packages including @radix-ui/react-checkbox, tailwindcss-animate, shadcn/ui, and lucide-react as requirements for implementation.
[DATA_EXFILTRATION]: The skill provides implementation examples using the fetch API to send data to relative backend endpoints (e.g., /api/teams/${teamId}/items/bulk-delete) for bulk processing.

bulk-select-actions