The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided text and documents (.docx) to generate academic proofreading reports, which creates a surface for indirect prompt injection and other data-based attacks. Besides that, the lack of sanitization could lead to Cross-Site Scripting (XSS) if the generated HTML reports are used to view malicious user-provided content.
Ingestion points: User-provided text and file content are ingested via scripts/init_run.py and scripts/word_converter.py.
Boundary markers: There are no explicit delimiters or instructions to the agent to treat the processed text as untrusted data.
Capability inventory: The skill can read and write files on the local system, convert documents to PDF, and generate visual HTML reports.
Sanitization: ABSENT. The AnnotationGenerator class in scripts/annotation_generator.py fails to sanitize or HTML-encode user-provided text before embedding it in the generated HTML report, creating a Cross-Site Scripting (XSS) vulnerability.
[EXTERNAL_DOWNLOADS]: The skill references several external Python dependencies for its linguistic analysis tasks, including languagetool-python, opencc, jieba, pyenchant, markdown, python-docx, and docx2pdf. While these are well-known and reputable libraries, the absence of a requirements.txt file in the skill package means exact versions are not pinned within the distribution.

content-proofreading