The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user and assistant messages to provide Hebrew translations and bidirectional alignment, which creates an attack surface for indirect prompt injection.
Ingestion points: Chat message content is ingested from the agent context via AssistantMessage and UserMessage rendering hooks in hooks/register.tsx.
Boundary markers: The translation logic in hooks/register.tsx passes raw message text directly to the $.model.complete call without using explicit delimiters or instructions to ignore embedded commands.
Capability inventory: The skill is authorized to invoke model completions ($.model.complete) and modify the rendered text of the conversation via ui.render hooks.
Sanitization: There is no sanitization or filtering of the message content before it is processed by the translation model, potentially allowing malicious instructions in the chat to influence the translation output.

hebrew-tools