Slide Deck From Doc

Overview

Reads any document (DOCX, PDF, or Markdown), extracts the logical structure and key content, then generates a professional slide deck with appropriate layouts, visual hierarchy, and speaker notes. Handles long documents by intelligently condensing content to fit the target slide count.

Workflow

<Definition - Slide Layout Types> Available layout archetypes for slide plan design:

<Definition - Style Profiles> Visual style applied to the presentation:

<Definition - Content Density Rules> Constraints for slide content to maintain readability:

• Max 6 bullet points per slide
• Max 12 words per bullet
• Max 3 levels of hierarchy on any single slide
• Speaker notes hold the full detail; slides hold the summary
• One key message per slide </Definition - Content Density Rules>

<Workflow - Convert description="End-to-end document to slide deck conversion." tools=[file_read_docx, file_read_pdf, file_read, pptx_plan, open_in_session_tab] triggers=["turn this into slides", "make a deck from this doc", "convert to presentation", "document to powerpoint", "create slides from this file"]

1. [Decide] Determine file type from extension:

   • .docx or .docm → use file_read_docx
   • .pdf → use file_read_pdf
   • .md or .txt → use file_read
   • Other → inform user that only DOCX, PDF, and Markdown are supported. Validate: File extension is recognized. If fails: Ask user to confirm the file path and format.

2. [Agent] Read the full document content. If the tool returns next_offset, continue reading with offset until complete. Validate: Content is non-empty and exceeds 50 characters. If fails: Inform user the file appears empty or unreadable.

3. [Think] Analyze the document structure:

   • Identify sections from headings, bold text, or logical breaks
   • Extract key points per section
   • Identify data/statistics worth highlighting as Key Stat slides
   • Identify quotes or callouts worth featuring
   • Determine the narrative arc and logical flow
   • Count total words to assess density Validate: At least 3 logical sections identified, OR fall back to paragraph chunking per Rule 4.

4. [Think] Design the slide plan using the following logic:

   • Assign Title/Cover as slide 1 (document title + subtitle/author if present)
   • If 4+ sections exist, add a Table of Contents slide
   • Assign Chapter Header for each major section transition
   • Map section content to appropriate layouts from Slide Layout Types
   • Place Key Stat/Quote slides for impactful data points
   • Use Comparison/Two-Column when contrasting ideas are present
   • End with Summary/Takeaways slide
   • Apply the selected style profile from Style Profiles
   • Ensure total slides do not exceed {{max_slides}}
   • Write speaker notes for each slide with full source detail Validate: Slide count is between 3 and {{max_slides}}. Every slide has a layout assignment.

5. [Agent] Present the slide plan using pptx_plan with the designed slides, applying the {{style}} theme. Validate: Plan renders in the design panel. If fails: Simplify layouts and retry.

6. [Ask user] Wait for user approval of the slide plan. If changes are requested, iterate on the plan and re-present. Validate: User explicitly approves (e.g., LGTM, looks good, approve, build it).

7. [Agent] Build the presentation via the canvas_pptx dependency following its build workflow with the approved plan. Validate: .pptx file is generated without errors in artifacts/. If fails: Retry with simplified layouts (fall back to Content + Bullets for problematic slides).

8. [Agent] Open the generated .pptx file in a session tab for user review using open_in_session_tab. Validate: File opens successfully. If fails: Provide the file path for manual opening.

</Workflow - Convert>