医学学术 PPT 自动化 (yixueAIganhuo-PPT)

What it does

This skill automates creating medical academic PowerPoint decks: it generates full‑slide 16:9 images using GPT Image 2 (or Codex built-in image_gen when available), extracts figures from PDFs, runs OCR (PaddleOCR v5 / PP-OCRv5) to reconstruct editable text layers, produces clean‑background images, and assembles two PPTX outputs — a non-editable image-only deck and a reconstructed editable deck with OCR text boxes and preserved figures. It enforces provenance rules so first-pass slides remain model outputs and runs the pipeline as an overlapped DAG for parallel generation and processing.

When to use it

Use this skill when you need to convert papers, reports, screenshots, or figure sets into presentation slides while preserving scientific figures and enabling editable text. It's ideal for medical/research slide generation where accurate figure handling, OCR reconstruction, and provenance of model outputs matter. Not suitable if you only need programmatic PowerPoint layout or simple templated slides.

What's included

• Scripts: pipeline scripts and OCR helpers are referenced in the skill (instructions for scripts/run_gpt_image2_slide.py, scripts/run_ocr_slide.py, scripts/make_clean_inputs.py, scripts/build_editable_pptx.py).
• References: extensive procedural rules and references for orchestration, provider config, and QA (mentioned in the skill body).
• Instructions: detailed intake flow (AskUserQuestion in Chinese), figure-count reconciliation for PDFs, concurrency and DAG rules, artifact provenance validation, and progress reporting hooks.

Compatible agents

Best used in environments that support script-backed GPT Image 2 providers or Codex subagent image_gen orchestration (Codex App/CLI when available). Works with local CLIProxyAPI, OpenAI/OpenRouter relays, and local PaddleOCR v5 for OCR processing.