The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes task titles and notes from external Google Tasks lists which are untrusted data sources.
Ingestion points: Task list data is ingested in handler.ts via googleClient.getTasksByListName(list_name) inside the getTaskList method.
Boundary markers: The skill lacks explicit boundary markers or instructions to the LLM to ignore potentially malicious instructions embedded within task items.
Capability inventory: The skill possesses the capability to modify user data through add_to_task_list and remove_from_task_list in handler.ts, creating a small risk if an injected instruction in a task note were to manipulate the agent into performing unwanted deletions or additions.
Sanitization: No sanitization or filtering is performed on the retrieved task titles or notes before they are returned to the agent context.

google-tasks