git:20260813.224f50f to git:20260928.c17f23d

1 added, 1 removed. Audit A to A.

# Contents
- `SKILL.md` โ€“ PDF ingestion workflow, completeness contract, and review loop.
- `references/` โ€“ guidance for reviewing OCR, conversations, and mixed-media images.
- `agents/openai.yaml` โ€“ user-facing skill metadata.
# Guidelines
- - Invoke the installed `kb` CLI; do not depend on a source checkout or implementation path.
+ - Invoke the installed `wordcell` CLI; do not depend on a source checkout or implementation path.
- Preserve the source PDF byte-for-byte, record sanitized remote URL provenance when present, and never persist its original absolute path.
- Keep native text, image text, and visual assets as independent evidence surfaces.
- Retain every extracted image even when its text is converted to Markdown.
- Preserve page, geometry, OCR method, confidence, and classification metadata.
- Never invent conversation authors, channels, timestamps, or reply relationships.
- Downgrade completeness when any page, image, tool, byte, time, or OCR boundary is reached.