Google Slides authoring

Coordinates are POINTS from the slide's top-left; a standard slide is 720x405 (read_slide_layout reports the real size). Everything is addressed by objectId from read_presentation. Pass requiredRevisionId from the latest read or write so a concurrent edit fails the write instead of corrupting it.

Building a new deck: compose, then look

Write each slide as a layout, not as coordinates. compose_slide takes a tree (stacks, grids, cards, text, images) or an HTML/CSS subset and lays the whole slide out in one call: equal peer sizes, even gutters, vertical centering, and contrast pairs hold by construction, text can fit: "shrink", and the reply carries every element's box plus lints on the plan. Per slide:

1. add_slide layout BLANK (or the deck's first slide with clear: true), then compose_slide with the SAME theme every time (palette roles background/surface/primary/textSecondary and the font family). Colors in the tree are roles, not hexes, so one theme change restyles a deck.
2. Read the reply's issues. text_overflow means shorten, fit: "shrink", or give the node more size; an overlap means two nodes share a box (usually a row that needs gap).
3. get_slide_thumbnail (LARGE) once per slide and LOOK at it: crowding, dead corners, a title that reads as body text, an image fighting the palette. Fix by re-composing the slide (delete the old elements first, or compose onto a fresh BLANK slide and delete_object the old one).

Tree recipes (column root; the root box is the slide minus a 48 pt margin):

• Title slide: stack column justify:center with a display text and an h2 subtitle in textSecondary, both align:center.
• KPI cards: h1, then a stack row size:150 gap:24 of cards, each with a caption label (textSecondary) over a kpi value.
• Text + image: stack row gap:32 of a body text (bullets:true, size:{flex:1}) and an image (size:{flex:1}); the image aspect-fits.
• Agenda: one body text with newline-separated items and bullets:true.

HTML form, same result: <section style="background:#0B2A4A; padding:48pt; gap:24pt"> with <h1>, <div style="display:flex; gap:24pt; height:150pt"> of <article class="card"><span class="caption">…</span><span class="kpi">…</span></article>, <ul><li> bullets, <img style="flex:1">. Only the documented tags and properties compile; anything else is refused by name, never dropped.

Editing an existing deck: place, inspect, fix, look

For decks that already have content, or elements the tree does not model:

1. insert_grid for any row, column, or grid of cards/text/images (one call, aligned, equal sizes); insert_text_box / insert_shape / insert_image for singletons.
2. read_slide_layout: every element's box, z-order, text style, estimated wrapped height (textFit.fits), and issues. Fix every error and warn (text_overflow, overlap, off_slide, low_contrast with the fill actually behind the text named) and the info items when cheap (near_miss_alignment, unequal_peers, tight_margin, empty_placeholder).
3. get_slide_thumbnail once the lints are clean, and look.

Grid conventions (when placing by hand)

Title at x 40, y 32, 640 wide, height 60-70 for one line of 28-34 pt (two lines need ~110); subtitle under it at 16-18 pt. Content region x 40, y 130, 640 x 235 (insert_grid's default). Footer at y 370, 12 pt. Keep 40 pt side margins unless an element is deliberately full-bleed. Process flows: insert_grid the step shapes (columns = steps), then connect_shapes between neighbors; connectors follow shapes when moved.

A 34 pt bold title wraps past ~32 characters in a 600 pt box and spills over whatever sits below (the API does not grow boxes). textFit in the layout read and issues in the compose reply predict this; long titles go to 24-28 pt, fit: "shrink", or two lines with the subtitle pushed down.

Palette and contrast

Pick a palette FIRST and reuse it on every slide (as compose_slide's theme.palette): background, one card/surface color, one accent, and a secondary text color; on-colors derive automatically. When the user names a template or brand, take its colors; otherwise these pairs pass WCAG AA as-is:

• Dark: background #0B2A4A, cards #1E4976, accent #4FC3F7, titles/text #FFFFFF, secondary #B8CCE0.
• Light: background #FFFFFF, cards #F1F3F4, accent #0B57D0 (white text on it), text #202124, secondary #444746.

Text color never defaults sensibly on a colored fill: every text-bearing call picks its color against the fill BEHIND the text (the card's, not the slide's). No mid-gray on mid-anything; on white, secondary text is #444746 or darker (#5F6368 is borderline, lighter fails). For a template or brand palette, run check_contrast on every text/fill pair you intend to use before writing (it returns a passing tint or shade for each failing pair); low_contrast in the layout read is the check afterwards.

Tables on colored slides render dark-on-dark from theme fills: set_table_cell_fill every cell (light fill + dark text, or accent + white) and restyle with set_text_style (cellRow/cellColumn). Prefer stat cards over a grid of numbers on title and summary slides; a table only for genuinely tabular data, with the header row filled in the accent and kept to a handful of rows.

Slides and repetition

• create_presentation starts with one blank slide (delete its placeholders, or fill them with set_shape_text); it cannot target a folder, so move it after with google-drive google_drive_move_or_rename.
• add_slide with a predefined layout fills title/body placeholders in the same call (fastest theme-consistent slide); move_slide reorders.
• Repeated layouts (one template, N variants): build the first slide fully, then per variant duplicate_object on the slide id (the copy lands right after) and replace_text scoped by pageObjectIds to the copy; distinct placeholders like {{TITLE}} keep the replaces unambiguous.
• set_shape_text REPLACES a shape's or cell's whole text (cell = {rowIndex, columnIndex}); set_text_style restyles all of one shape or cell, not a range. update_element_position moves but does not resize: delete and re-create at the right size instead.
• Speaker notes: set_shape_text with pageObjectId = the slide's notesPageId and objectId = speakerNotesObjectId from read_presentation.
• Whole-deck render: google-drive google_drive_export_file (pdf).