The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, specifically monday.com boards, item names, and comments, and uses this data to inform actions taken via write-capable tools.
Ingestion points: Untrusted data enters the agent context through get_board_items_page, get_board_info, and search results in SKILL.md.
Boundary markers: The instructions lack explicit boundary markers or delimiters (e.g., XML tags or backticks) to isolate external data from internal instructions, nor do they include warnings to ignore instructions embedded in board content.
Capability inventory: The skill has significant capabilities to modify the environment, including create_item, change_item_column_values, create_items, update_items, and all_monday_api (for GraphQL mutations) as documented in SKILL.md.
Sanitization: There is no mention of sanitizing, escaping, or validating the content of strings retrieved from the board before they are processed by the agent or included in subsequent API calls.

task-management