The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data which serves as a potential vector for indirect prompt injection.
Ingestion points: According to SKILL.md, the agent ingests data from user-provided files, pasted third-party content, screenshots, tables, and benchmark samples.
Boundary markers: The skill explicitly references ../eva-shared/references/shared/06_external-material-safety_外部材料安全边界.md to be loaded specifically when handling external materials to define safety boundaries.
Capability inventory: The skill can read local reference files, transfer execution to other modules within the Eva ecosystem, and perform file system operations such as packaging data to the desktop.
Sanitization: It includes an authenticity check via ../eva-shared/references/quality/00_eva-ai-check_表达真实性审查.md to evaluate the nature of processed text.
[COMMAND_EXECUTION]: The skill instructions in SKILL.md provide triggers for managing local data, such as "export Eva data", "backup all Eva memory cards", and "package Eva data to desktop" (把 Eva 数据打包到桌面). These features involve the agent performing file system operations and potentially executing shell commands to archive or move data to the user's local environment.

eva-think