The Agent Skills Directory

[SAFE]: The skill enforces a strict interactive approval process for all 19 checks. The AI is explicitly instructed to wait for user confirmation ('yes/no/edit') for every issue before using the Edit tool, which serves as a primary defense against unauthorized modifications.\n- [SAFE]: Access is scoped to academic artifacts (LaTeX files, BibTeX files, and PDFs) within a user-provided directory. The skill does not attempt to access sensitive system paths or credentials.\n- [SAFE]: Analysis of the skill instructions and the secondary Chinese prompt file found no patterns of obfuscation, remote code execution, or data exfiltration. The use of tools like Bash and WebFetch is aligned with academic research and verification tasks.

paper-polish