The Agent Skills Directory

[PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection as it ingests untrusted data from email subjects and snippets.
Ingestion points: src/index.ts (within the classifyEmail and classify methods).
Boundary markers: Absent; untrusted email content is processed directly by the scoring algorithm without the use of delimiters or 'ignore' instructions.
Capability inventory: The skill can archive, trash, and mark emails as read through bulkArchive, bulkMarkAsRead, and bulkTrash functions in src/index.ts.
Sanitization: Content is converted to lowercase for comparison but is not sanitized against adversarial patterns intended to manipulate classification logic or downstream agent actions.

inbox-triage