The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection because it ingests data from external String Catalog files (.xcstrings) and user-provided terminology or style instructions, which are then used to generate prompts for sub-agents. If these external sources contain malicious instructions, they could be executed by the sub-agents during the translation process.
Ingestion points: Untrusted data enters the agent context through the StringCatalogRead tool, which retrieves keys and strings from project files, and through the user's provided "terminology and style choices."
Boundary markers: The delegation prompt template uses code blocks for keys but lacks robust delimiters or "ignore instructions" markers to isolate the content of the translations and style guidance from the sub-agent's core instructions.
Capability inventory: The skill can modify project configurations using LocalizationPlanner and invokes the xcode-skills:translation skill in sub-agents, which typically involves file-writing capabilities.
Sanitization: The skill is explicitly instructed to forward string keys and user guidance "exactly" or "verbatim" as received, without sanitizing or filtering for potential embedded prompt injection payloads.

translation-coordinator