The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the subprocess.run function to execute LaTeX compilers like xelatex and citation tools like bibtex to verify the generated document structure. These operations are restricted to specific, predefined commands and occur within the project's runtime environment.- [DATA_EXFILTRATION]: Snippets of user code, reference metadata, and document text are transmitted to LLM providers (Anthropic or OpenAI) for thematic analysis and narrative synthesis. This transmission is central to the skill's generative functionality.- [PROMPT_INJECTION]: The skill processes project-sourced data (source code, LaTeX, and BibTeX) which presents an indirect prompt injection surface. Maliciously crafted content in project files could theoretically influence AI-generated narratives, though the skill's output is sanitized and structurally validated by the integrated FormatGuard.- [DATA_EXPOSURE]: The skill scans and reads local project files including images, source code, and bibliographies to facilitate content generation. Path traversal protections are implemented in the skill_controller.py and format_guard.py modules to ensure access is restricted to the target project directory.

complete-example