The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted repository content (documentation files) as its primary input, which creates a surface for indirect prompt injection where malicious instructions embedded in the documents could influence agent behavior.
Ingestion points: Documentation files (.md) being translated or updated, and the repository's terminology table.
Boundary markers: The skill lacks explicit boundary markers or instructions to treat documentation content as data only, increasing the risk that the agent might follow instructions found within the text.
Capability inventory: The skill is instructed to execute local shell scripts (translation-brief, pairing gate) and perform file system write operations to update documentation pairs and consistency records (.i18n.yaml).
Sanitization: No sanitization or validation logic is defined for the content read from the repository documents before processing or interpolation into subagent prompts.

translate-docs