The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a custom script scripts/parse_file.py to parse documents. This script is invoked using the [INVOKE:parse_file(...)] pattern. The script is designed to process specific file extensions (.pdf, .docx, and common image formats) and includes error handling for unsupported file types, ensuring it operates only on intended file types within the sandbox environment.
[PROMPT_INJECTION]: The skill processes external document content, which creates a surface for indirect prompt injection. To mitigate this, the skill's generation prompts in references/generation-prompts.md use clear delimiters and include explicit 'strong constraints' (如【资料使用强约束】) that command the AI to rely exclusively on the provided text and maintain its professional persona. These safeguards reduce the risk of instructions within a document overriding the agent's behavior.
[SAFE]: All identified dependencies (PyMuPDF, python-docx, and Pillow) are well-known, legitimate libraries for file and image processing. The skill's operations are confined to reading user-provided files and writing to a local output directory, with no hardcoded credentials, network exfiltration patterns, or persistence mechanisms detected.

file-to-article-generator