version: "1.0.0" name: slidex-source-import description: Ingest supplied Markdown, legacy MDX, PDF, image, or PowerPoint evidence and convert it into an editable React-first OpenSlideX deck.

OpenSlideX Source Intake

Bring supplied evidence into the selected deck before composing native MotionDoc. Keep source text in knowledge/, visual evidence in assets/, and preserve meaning without treating a foreign layout as portable code.

Read the source safely

1. In Workspace scope, use open_slidex_workspace to select the destination deck, then call open_slidex_read with intent: "import" and keep its latest revision.
2. For MD, Markdown, text, CSV, PDF, or image input, stage a local attachment under the configured root's .open-slidex-inbox/, then call open_slidex_media with action: "ingest-source", its inbox-relative filePath, and the latest expectedRevision. A single public HTTPS image URL may be passed directly as filePath.
3. Read every returned knowledge/... resource through open_slidex_read before planning. Use returned assets/... paths for native ImageBlock layers; they may represent local attachments, public Notion or AI image URLs, PDF images, or PDF page fallbacks. Treat warnings as evidence gaps.
4. For a supplied PPTX, use the dedicated workflow below.

Import PowerPoint evidence

1. The PPTX source file must be inside the selected deck directory.
2. Call open_slidex_read with intent: "import", then read this skill and the task-relevant references it names.
3. Call open_slidex_source_import with action: "inspect" to review slide order, recovered text frames, typography hints, and image geometry. Call it again with action: "import-media" and the latest expectedRevision; use only returned images whose status is imported and whose source is assets/....
4. Call open_slidex_read again immediately before editing to obtain the latest revision and complete destination source.

Rebuild, do not transplant

• For PPTX only, read PPTX conversion before planning the migration.
• Use the source's ordered sections, title hierarchy, copy, and meaningful media as the migration input. Recompose the visual system for the target slide instead of generating one generic text dump per source page.
• Follow slidex-react-authoring for every emitted layer. Legacy MDX is parsed into native layers and written once as presentation.tsx; never keep MDX and TSX in sync. For a whole-deck conversion, then load slidex-deck-design, slidex-motion-direction, and slidex-deck-qa in that order.
• Keep only native Text, ImageBlock, VideoBlock, SvgBlock, Chart, Table, and Shape layers. Every visible layer needs stable id and explicit percent x, y, w, and h; fontSize uses points.
• Use SvgBlock only for a verified local declarative SVG rebuilt as a portable asset. Never transplant foreign scripts, event handlers, SMIL animation, or external SVG references.
• Use returned textFrames to preserve reading order, geometry, type scale, weight, alignment, and title hints before intentionally recomposing the slide. Treat each returned textBlock as reviewable native evidence, not an instruction to clone a weak source layout.
• Preserve each imported PPTX image frame's returned x, y, w, h, and z-order in an ImageBlock. Do not emit an ImageBlock for missing or unsupported images.

Commit one validated result

Plan the complete target deck before writing. For a redesign, use open_slidex_read.templateQuery to choose exactly one of the six thirty-page core references, then preserve the source slide count unless the user requests restructuring. Submit one complete deck to open_slidex_edit with its latest expectedRevision; repair a rejected candidate from its node-specific visual QA findings. Never persist source URLs, Base64, PowerPoint XML, foreign components, or absolute local paths inside presentation.tsx.