okf-guard: a content-safety scanning layer for OKF pipelines

[darshanNhb]

Sharing a tool that might be useful to others building OKF pipelines: okf-guard, a Python library that screens raw source content (PDF, DOCX, PPTX, XLSX, HTML) for hidden text and prompt-injection-style phrasing before it's written into an OKF concept file.

The motivation: OKF bundles are increasingly auto-generated from sources of mixed trustworthiness (scraped pages, PDFs, exports), and nothing in the current tool ecosystem checks the content for anything adversarial — only structural validity. If someone hides an instruction in white-on-white PDF text, a hidden spreadsheet row, or an invisible speaker note, it can end up sitting in a "trusted" concept file indistinguishable from real content.

It's meant to plug into an existing generator's extraction step via a single sanitize() call, not to replace any generator. Fully local/deterministic in this release (no LLM calls, one hard dependency), targets the v0.2 schema — writes generated/sources, never a self-asserted verified.

Repo: https://github.com/darshanNhb/okf-guard
PyPI: pip install okf-guard

It was recently reviewed and added to the community tools directory at okf.md/tools — happy to answer questions or take feedback here too.