PaddleOCR Document Parsing

**Overall verdict:** Strong, specialized skill for rich document OCR and layout parsing, best suited for complex PDFs and images where tables, formulas, and multi-column layouts matter. It is tightly scoped, well-documented, and enforces correct use of the PaddleOCR API through a single script entrypoint. **Strengths** - High-fidelity structural extraction (text, tables, formulas, figures, layout, reading order) with a clean JSON envelope. - Clear operational contract: always use `python scripts/vl_caller.py`, never fall back to in-model vision, and always return complete content as requested. - Good guidance for configuration, API errors, and large-file handling, including page-range splitting. **Key risks / constraints** - Hard dependency on an external PaddleOCR API; failures (misconfiguration, quota, network) cannot be mitigated by the agent and must be resolved by the user. - Requires prior environment configuration for API URL, token, and optional timeout; initial setup may be non-trivial for non-technical users. - All documents are sent to a third-party OCR service, which may be a concern for sensitive or regulated content. **Recommended scenarios** - Complex document ingestion pipelines where you need full structure (tables, formulas, layout) rather than plain OCR text. - Back-office or research workflows processing invoices, financial statements, scientific papers, and multi-column PDFs. - As a dedicated OCR/layout microservice within a larger OpenClaw agent system, where another skill or agent performs downstream summarization, analysis, or search on the structured JSON output.