Local conversion, your files stay with you

Scope

The main security surface is:

• Python extraction code that parses document files you provide
• Optional dependencies installed on request (pip install … when you choose --install-missing yes)

If your agent's model runs in the cloud, text you feed it follows that provider's normal data terms, same as any prompt. The converter itself does not upload sources.

Supported versions

The latest released 1.x version receives fixes. Reproduce issues against the most recent tag before reporting.

Reporting a vulnerability

Do notopen a public issue for a security problem. Use GitHub's private vulnerability reporting:

• Go to the repository's Security tab → Report a vulnerability

Include affected version, a minimal reproduction (ideally a small sample file or crafted input), and the impact you observed. Maintainers aim to acknowledge within a few days.

Good practices for users

• Run python3 scripts/extract.py --check to see which extractors are in use. Install dependencies yourself if you prefer to control what is added.
• Only convert documents you trust and have the right to process.
• Keep generated skills of copyrighted third party books private.

Recent hardening (v1.3.0)

• Advisory prompt injection scan on generated skills
• Invisible Unicode scrubbing during extraction
• DOCX XXE and entity expansion rejection
• Subprocess argument injection hardening for external tools
• Dependency CVE review on pull requests