The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and process content from external document files (.typ and .qmd) which are not under the system's control.
Ingestion points: Data is read from files specified by the source_file argument using the Read, Grep, and Glob tools.
Boundary markers: The skill lacks explicit instructions to treat document content as data only or to ignore instructions embedded within the source documents.
Capability inventory: The skill is granted access to high-privilege tools including Bash, Task, Write, and Edit, which could be abused if the agent inadvertently follows instructions found inside a document being translated.
Sanitization: There are no described methods for sanitizing or validating the document content before it is processed by the agent or passed to the allowed tools.

translate