The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell command execution for standard academic document processing and compilation.
Evidence: Appendix D details the use of the exec tool to run xelatex, biber, and bibtex for LaTeX document generation.
Evidence: Appendix E and G define usage patterns for the exec tool to perform document conversions via pandoc (e.g., markdown to docx/pdf).
Evidence: Appendix E mentions executing local scripts, including a Node.js script (md2pdf.js) and a shell script (setup-env.sh), for the PDF export pipeline located in the skills/md2pdf-export/ directory.
[DATA_EXFILTRATION]: All data management operations are confined to the agent's defined workspace and local library environments.
Evidence: The skill uses workspace_save, workspace_export, and library_export_bibtex to manage outputs within outputs/drafts/ and outputs/exports/. No unauthorized external network requests or data transfer patterns were observed.
[PROMPT_INJECTION]: The skill contains no instructions designed to bypass agent safety guidelines or override system behavior.
Evidence: The "CRITICAL" and "MUST" instructions are used exclusively to enforce the academic writing protocol and iterative review phases. No "ignore previous instructions" or similar adversarial patterns are present.
[REMOTE_CODE_EXECUTION]: The skill does not perform remote downloads or execute code from untrusted external sources.
Evidence: All script executions and tool invocations refer to local binaries or scripts within the skill's own environment.

Writing SOP