The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs local file system write operations to persist user translation preferences. These files are stored in a path derived from user-controlled project and file names: {项目根目录}/wenqu-skills/{文件名}/references/preferences.md. While this is functional, it requires the underlying agent to handle the filename safely to prevent path traversal.
[DATA_EXFILTRATION]: The skill reads translation guidelines and review criteria from a peer skill's directory (~/.agents/skills/wenqu-review/references/r2-translation.md). This cross-directory access is documented as part of the vendor's internal skill suite integration.
[PROMPT_INJECTION]: The skill ingests untrusted text from documents, papers, and codebases for translation. This creates an indirect prompt injection surface where instructions hidden in the source text could attempt to manipulate the agent's behavior. The skill's analysis-and-rewrite methodology serves as a functional barrier against simple instruction-based injections.

wenqu-translate