The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The main SKILL.md file (§1.0) contains instructions for the AI agent to auto-update the skill by executing npx skills add pointing to a different GitHub repository (tfboy1/academic-paper-writer) than the current skill's author (BoHeFan/academic-paper-writer-pro-2). This introduces a supply chain risk where the agent could be tricked into downloading and running code from an unverifiable source.
[COMMAND_EXECUTION]: The script docx/scripts/office/soffice.py dynamically writes a C shim to the system's temporary directory and compiles it into a shared library using gcc. This library is then injected into the execution environment of other processes using the LD_PRELOAD variable to intercept system calls like socket and accept. This is a highly invasive technique typically used for bypassing security sandboxes.
[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run to execute system-level binaries including soffice (LibreOffice), pandoc, mmdc (Mermaid CLI), git, and gcc. These commands are executed with arguments derived from document content and configuration files.
[EXTERNAL_DOWNLOADS]: The skill provides numerous download links in templates/README.md for official conference templates (IEEE, ACM, APA, Springer). While these specific links point to well-known academic domains, the skill's instructions encourage the agent to download and process arbitrary binary files (.docx, .pdf) from these external locations.
[COMMAND_EXECUTION]: In SKILL.md (§3.3), the Mermaid diagram generation pipeline explicitly disables Chromium's security sandbox by passing the --no-sandbox and --disable-setuid-sandbox arguments to mmdc (Puppeteer), which increases the risk if processing untrusted or malicious Mermaid code.

academic-paper-writer-pro-2