powerpoint-automation

PowerPoint Automation generates and edits professional PPTX presentations from web articles, blog posts, existing PPTX files, or templates, using an Orchestrator-Workers pattern. It fits tasks such as turning an article into slides, translating a deck between English and Japanese, editing a currently open PowerPoint file via COM Automation, or running RefURL, overflow, notes, and hyperlink review passes.

The design centers on content.json as the single source of truth (SSOT); extraction, translation, generation, and review all read and write against it so that each phase stays separated. The pipeline runs TRIAGE, PLAN, PREPARE_TEMPLATE, EXTRACT, TRANSLATE, BUILD, REVIEW, and DONE, with a human confirmation step in PLAN. Supporting Python scripts include extract_images.py (source to content.json), create_from_template.py (content.json to PPTX using a template and layout config), reconstruct_analyzer.py (existing PPTX back to content.json), and validators validate_content.py and validate_pptx.py (schema and overflow checks). Three agents divide the work: an Orchestrator for coordination, a Localizer for EN/JA translation, and a PPTX Reviewer for final quality.

A large set of operating rules encodes hard-won practices: never overwrite an open PPTX with python-pptx (lock first, use COM for open decks), keep operational and internal text in speaker notes rather than on customer-facing slides, build architecture diagrams from shapes instead of ASCII art, verify Azure/Microsoft technical content against primary sources, and perform rendered-image visual QA to catch overflow, tiny fonts, template drift, and stale wording before handoff. Extensive reference files document scripts, use cases, content guidelines, COM Automation instructions, and template handling. The skill is licensed CC BY-NC-SA 4.0.