The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/export_rich.py uses subprocess.run to invoke pandoc for document format conversion. The implementation follows security best practices by passing arguments as a list rather than a shell string, effectively preventing command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill identifies Pandoc and Playwright as system-level dependencies. It does not attempt to download or install these automatically; instead, it provides clear error messages instructing the user to install them through official channels, ensuring transparency and user control.
[DATA_EXPOSURE]: The SKILL.md file contains explicit privacy guidelines that instruct the agent to distinguish between internal context (like names or prompt history) and public content. It mandates that only user-approved fields enter the final export, which is a strong proactive measure against data leakage.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted Markdown and HTML data for processing. While the transformation to static formats (like PDF/DOCX) naturally limits the impact of embedded instructions, the script also uses html.escape when building media tags from manifest data to mitigate HTML injection risks in the generated outputs.

lov-rich-export