The Slides Agent is a professional presentation designer that converts topics, documents, or raw content into polished, visual HTML-based slide decks — then exports them to fully editableDocumentation Index
Fetch the complete documentation index at: https://mintlify.com/UnkleFunk/HouseMusicSwarm-/llms.txt
Use this file to discover all available pages before exploring further.
.pptx files. Each slide is authored as standalone HTML with shared theme CSS, so the pipeline preserves Google Fonts, CSS gradients, inline SVGs, and custom layouts with pixel-perfect fidelity when rendered to PowerPoint via Playwright and dom-to-pptx. The agent always clarifies scope before generating, plans the full deck narrative, researches content in parallel, and only starts building slides once all assets and facts are gathered.
Tools
InsertNewSlides
Inserts blank slide placeholders at a specified position in the deck. Accepts a task brief and page count, then returns an outline and execution plan with template keys. Always run this before
ModifySlide when adding new slides.ModifySlide
Generates or edits a slide’s HTML by dispatching a fully isolated sub-agent with the task brief as its only context. Automatically returns a screenshot after every modification so the agent can inspect for overflow, broken layouts, or missing images.
ManageTheme
Creates or overwrites the shared
_theme.css file for a project. Stores CSS variables for the color palette, typography, spacing, and reusable layout classes (cards, labels, glass panels). All slides link this file to stay visually consistent.ReadSlide
Returns the raw HTML source of a named slide file. Use this to inspect the current design or content of a slide before making targeted edits, without triggering a full re-render.
SlideScreenshot
Renders a single HTML slide to a JPEG image attachment using Playwright at 1280×720. Use this to visually inspect slides that weren’t just modified (ModifySlide screenshots automatically — this is for consistency checks on other slides).
CheckSlide
Renders a single HTML or PPTX slide to an image and returns it as a file attachment. Useful for inspecting rendered output at any point, including PPTX slide previews using LibreOffice.
CheckSlideCanvasOverflow
Detects content that overflows the slide canvas by adding grey padding around each slide, rendering to image, and inspecting the padding margins for non-grey pixels. Returns the failing slide indices and debug images showing the overflow. Requires LibreOffice.
BuildPptxFromHtmlSlides
Converts an ordered list of HTML slides into a fully editable
.pptx file. Uses Playwright to measure every element’s exact computed position and style, then maps them to native PPTX shapes. Custom fonts are auto-embedded. Output is auto-versioned: my_deck.pptx, my_deck_v2.pptx, etc.RestoreSnapshot
Rolls back the working slides to any previous export. Reads HTML from the snapshot directory created alongside each
.pptx export, restores slide_01.html, slide_02.html, etc., and re-extracts the theme CSS.CreatePptxThumbnailGrid
Converts all slides in a PPTX to images and arranges them in a grid layout (up to 5 columns by default). Useful for a fast visual overview of an entire deck. Automatically creates multiple grid files for large presentations. Requires LibreOffice and Poppler.
ImageSearch
Searches for stock images across Unsplash, Pexels, and Pixabay using a text query. Returns image URLs, descriptions, dimensions, and photographer credits. Use this for real-world photos, textures, and backgrounds — not for brand-specific logos or icons.
DownloadImage
Downloads an image from a URL into the project’s
assets/ folder. Validates that the response is a real image (not an HTML error page). Use the downloaded path as ./assets/{filename} in slide HTML — the PPTX exporter auto-embeds it as base64.GenerateImage
Generates AI images using Google Gemini. Two modes:
diagram (uses gemini-3-pro-image-preview, optimized for flowcharts, org charts, pyramids with text) and concept_art (uses gemini-2.5-flash-image, faster generation for illustrations and atmospheric images). Saves to the project’s assets/ folder.EnsureRasterImage
Converts vector and container formats (SVG, EMF, WMF, HEIC, PDF, EPS) to raster PNG using Inkscape, Ghostscript, or ImageMagick. Already-raster files are returned as-is. Use this before embedding any non-raster asset in a slide.
ExtractPptxTextInventory
Extracts all text shapes from a PPTX file into a structured JSON file, organized by slide. Includes position, dimensions, placeholder types, paragraph formatting, and overflow/overlap issue detection. Use before batch text editing.
ApplyPptxTextReplacements
Applies batch text replacements to an existing PPTX file from a JSON file following the inventory structure. Preserves all existing formatting. Validates that referenced shapes exist and that overflow does not worsen after replacements.
RearrangePptxSlidesFromTemplate
Reorders the slides in a PPTX file according to a provided sequence of slide indices. Use this to restructure an existing deck without rebuilding from HTML.
CreateImageMontage
Creates a labeled grid montage from a collection of images (supplied as a list of paths or a directory). Configurable column count and cell width. Non-raster formats are auto-converted before inclusion. Useful for reviewing slide image assets or creating visual indexes.
DeleteSlide
Deletes an HTML slide file from a presentation project. Accepts either a specific slide name (e.g.
slide_03) or a list of 1-based slide indexes. Use this to remove slides that are no longer needed after restructuring or replacing a section of the deck.Image sources
TheImageSearch tool queries up to three stock photo providers. Set whichever keys you have — the agent skips any provider whose key is missing and proceeds with the rest.
| Provider | Environment variable |
|---|---|
| Pexels | PEXELS_API_KEY |
| Pixabay | PIXABAY_API_KEY |
| Unsplash | UNSPLASH_ACCESS_KEY |
GenerateImage tool uses Google Gemini for AI image generation and requires GOOGLE_API_KEY.
Example prompts
- “Create a 10-slide investor pitch for OpenSwarm”
- “Add a title slide and agenda to this deck”
- “Change all slides to use a dark theme with the company brand colors”
- “Turn this PDF document into a professional slide deck”
- “Polish the layout — the content slides look too text-heavy”
Notes and requirements
Anthropic’s Claude produces significantly better slide HTML output than other providers. For the best results, set your
DEFAULT_MODEL to a Claude Sonnet model (e.g. anthropic/claude-sonnet-4-6). The HTML writer sub-agent inside InsertNewSlides and ModifySlide also defaults to Claude when available.