Skip to content

Cursor plugin specification and official plugins

TypeScript

Update bundled skills

Jacob Sampsone2ad3f224a32
diff --git a/third_party/google-docs/.cursor-plugin/plugin.json b/third_party/google-docs/.cursor-plugin/plugin.json
index b43c5ad5..dbe8280a 100644
--- a/third_party/google-docs/.cursor-plugin/plugin.json
+++ b/third_party/google-docs/.cursor-plugin/plugin.json
@@ -1,7 +1,7 @@
{
"name": "google-docs",
"displayName": "Google Docs",
- "version": "1.0.0",
+ "version": "1.1.0",
"minClientVersions": {
"cursor": "3.22.0",
"grokbot": "0.30.0",
diff --git a/third_party/google-docs/CHANGELOG.md b/third_party/google-docs/CHANGELOG.md
index 59e90831..168b4f36 100644
--- a/third_party/google-docs/CHANGELOG.md
+++ b/third_party/google-docs/CHANGELOG.md
@@ -2,6 +2,10 @@
All notable changes to this plugin will be documented here.
+## 1.1.0
+
+- Refreshed the `google-docs` skill for the latest tool surface.
+
## 1.0.0 - initial release
- Logo: official Google Docs product icon on a padded white tile.
diff --git a/third_party/google-docs/skills/google-docs/SKILL.md b/third_party/google-docs/skills/google-docs/SKILL.md
index 7afe4f25..4885a5ad 100644
--- a/third_party/google-docs/skills/google-docs/SKILL.md
+++ b/third_party/google-docs/skills/google-docs/SKILL.md
@@ -1,23 +1,37 @@
---
name: google-docs
-description: Author and edit Google Docs through the google-docs MCP tools - structured writing, character and paragraph styling, lists, tables, images, and safe index math. Use when creating a formatted document, editing document content, or applying styles.
+description: Author and edit Google Docs with the google-docs tools - structured writing, character and paragraph styling, lists, tables, images, headers/footers. Use when creating a formatted document, editing document content, or applying styles.
---
# Google Docs authoring
-## The index contract (the #1 source of bugs)
-
-Docs edits address zero-based UTF-16 index ranges within one tab. Every write
-SHIFTS later indexes, so:
-
-1. `read_document` first; it returns each element's `{start, end}` plus the
- `revisionId`.
-2. Compute ONE edit from that read, pass `requiredRevisionId`, apply.
-3. Re-read before the next positional edit. Never reuse indexes across
- writes, and never do arithmetic across two writes.
-
-Order multi-part positional work BACK-TO-FRONT (highest index first) when
-re-reading between steps is too expensive; earlier ranges stay valid.
+## Locate text, never compute indexes (the #1 source of bugs)
+
+Docs edits address zero-based UTF-16 index ranges within one tab, and every
+write SHIFTS later indexes. Guessed or stale indexes silently hit the wrong
+paragraphs - the classic symptom is two lists merging into one continued
+numbering. The tools remove the need for index math:
+
+- `find_text` returns FRESH `{start, end}` ranges for exact text - the
+ direct source of indexes for `apply_text_style`, `apply_paragraph_style`,
+ `delete_content`, and `set_paragraph_format`. Locate, then edit, in
+ back-to-back calls.
+- `create_list` takes `firstItemText`/`lastItemText` anchors and resolves
+ the range itself: name the first and last item paragraphs and no index
+ ever touches your hands. Use one `create_list` per list with the right
+ preset (BULLET_* for bullets, NUMBERED_* for numbers); never stretch one
+ call across two visually separate lists.
+
+When a range must come from structure instead of a phrase (styling a whole
+paragraph, addressing a table), read it from `read_document` elements: a
+paragraph's `{start, end}` is the range for paragraph-level tools. END IS
+EXCLUSIVE: one short misses the last character, one long styles the
+character after the phrase (the classic italic-or-link off-by-one).
+
+When raw indexes are unavoidable: `read_document` first, compute ONE edit,
+pass `requiredRevisionId`, apply, and re-read before the next positional
+edit. Order multi-part positional work BACK-TO-FRONT (highest index first)
+when re-reading between steps is too expensive; earlier ranges stay valid.
## Creating a document
@@ -31,6 +45,9 @@ re-reading between steps is too expensive; earlier ranges stay valid.
revision-guarded automatically. `paragraphStyle` styles the new paragraphs
and `bold`/`italic` style the inserted text in the same call, so a
well-structured document is a sequence of appends with no styling pass.
+- Doc creates cannot set a parent folder. When placement matters, create
+ and then google-drive `google_drive_move_or_rename` with
+ `destination_folder_id`.
## Structure is not optional
@@ -52,17 +69,40 @@ afterwards.
line spacing (percent, 100 = single).
- Lists: `create_list` over the paragraphs' range; each newline-separated
paragraph becomes one item. Insert the plain lines first, then bullet them.
+- The preset is part of correctness: enumerations get a BULLET_* preset by
+ default; NUMBERED_* only when the content is ordered (steps, rankings) or
+ the request says numbered.
+- Before bulleting, check the read for paragraphs that ALREADY show a
+ leading bullet - text appended directly after a list inherits its
+ membership, and `create_list` over such paragraphs restyles the ORIGINAL
+ list, continuing its numbering (a second list showing 4, 5, 6). Strip
+ first: `create_list` with `remove: true` over exactly the new paragraphs,
+ then apply the wanted preset.
## Tables and images
- `insert_table` inserts an EMPTY grid. To fill it: re-read (cells appear
with their own index ranges) and `insert_text` into cells from the LAST
cell to the first, so earlier cell indexes stay valid.
+- Style a table AFTER filling it, addressing it by the table element's
+ `start` from `read_document`: `set_table_cell_shading` the header row
+ (pick the row's text color against that fill), `merge_table_cells` for a
+ title cell spanning columns (top-left content survives, the rest is
+ dropped), `set_table_column_width` for fixed widths in points. Structural
+ edits (merge) shift indexes: re-read before further text edits.
- `insert_image` needs a PUBLICLY fetchable URL (Google fetches it; auth
- walls and localhost fail), PNG/JPEG/GIF. Size with width/height in points
- (aspect kept when only one is set).
+ walls and private hosts fail), PNG/JPEG/GIF. Size with width/height in
+ points (aspect kept when only one is set).
- `insert_page_break` starts a new page at an index.
+## Headers and footers
+
+`set_header_footer` sets the default header or footer in one call: it
+creates the segment when the doc has none, replaces the text otherwise,
+and empty text clears it. The text is static (no page-number fields exist
+in the API); it applies to the first tab. Never `insert_text` a fake header
+into the body - it scrolls away with the content.
+
## Bulk text ops
`replace_text` replaces literal strings everywhere (or in named tabs) with
@@ -70,8 +110,10 @@ no index math; prefer it for renames and corrections. `delete_content`
rejects ranges that split table structure or surrogate pairs; deleting a
paragraph's final newline merges it with the next.
-## Verifying
+## Verify your work
Re-read with `read_document` for structure and styles, or google-drive
`google_drive_read_file` for the whole doc as Markdown, or
-`google_drive_export_file` (pdf) when the visual result matters.
+`google_drive_export_file` (pdf) when the visual result matters. After any
+styling call, confirm the styled span covers exactly the intended text
+before moving on.
diff --git a/third_party/google-sheets/.cursor-plugin/plugin.json b/third_party/google-sheets/.cursor-plugin/plugin.json
index b1e3e8dc..ebee059c 100644
--- a/third_party/google-sheets/.cursor-plugin/plugin.json
+++ b/third_party/google-sheets/.cursor-plugin/plugin.json
@@ -1,7 +1,7 @@
{
"name": "google-sheets",
"displayName": "Google Sheets",
- "version": "1.0.0",
+ "version": "1.1.0",
"minClientVersions": {
"cursor": "3.22.0",
"grokbot": "0.30.0",
diff --git a/third_party/google-sheets/CHANGELOG.md b/third_party/google-sheets/CHANGELOG.md
index 9ec31b18..fb421ace 100644
--- a/third_party/google-sheets/CHANGELOG.md
+++ b/third_party/google-sheets/CHANGELOG.md
@@ -2,6 +2,10 @@
All notable changes to this plugin will be documented here.
+## 1.1.0
+
+- Refreshed the `google-sheets` skill for the latest tool surface.
+
## 1.0.0 - initial release
- Logo: official Google Sheets product icon on a padded white tile.
diff --git a/third_party/google-sheets/skills/google-sheets/SKILL.md b/third_party/google-sheets/skills/google-sheets/SKILL.md
index de65b591..20177588 100644
--- a/third_party/google-sheets/skills/google-sheets/SKILL.md
+++ b/third_party/google-sheets/skills/google-sheets/SKILL.md
@@ -1,6 +1,6 @@
---
name: google-sheets
-description: Build and edit Google Sheets through the google-sheets MCP tools - values, formatting, charts, conditional formatting, structure, and validation. Use when creating a spreadsheet or dashboard, formatting cells, adding charts, or reorganizing tabs and rows.
+description: Build and edit Google Sheets with the google-sheets tools - values, formatting, charts, conditional formatting, structure, validation, named and protected ranges. Use when creating a spreadsheet or dashboard, formatting cells, adding or fixing charts, or reorganizing tabs and rows.
---
# Google Sheets authoring
@@ -10,16 +10,17 @@ description: Build and edit Google Sheets through the google-sheets MCP tools -
- VALUE tools (`read_range`, `write_range`, `append_rows`, `clear_range`)
address a TAB TITLE plus A1 cells: `{sheet: "Data", cells: "A1:B4"}`.
- STRUCTURE/FORMAT tools (`format_cells`, `set_borders`, `merge_cells`,
- `add_chart`, `add_conditional_formatting`, `insert/delete_rows_or_columns`,
- `freeze_rows_or_columns`, `sort_range`, `set_data_validation`,
- `set_basic_filter`, `add_banding`, `set_column_width`,
- `rename/delete_sheet_tab`) address the numeric `sheetId`.
+ `add_chart`, `add_conditional_formatting`, `insert_rows_or_columns`,
+ `delete_rows_or_columns`, `freeze_rows_or_columns`, `sort_range`,
+ `set_data_validation`, `set_basic_filter`, `add_banding`,
+ `set_column_width`, `rename_sheet_tab`, `delete_sheet_tab`) address the
+ numeric `sheetId`.
`get_spreadsheet` returns both (each tab's title AND sheetId) with no cell
data - call it first and keep the mapping. A new tab's sheetId is in
`add_sheet_tab`'s reply (`replies[0].addSheet.properties.sheetId`).
-## Sequencing a styled sheet
+## Sequencing a styled sheet: values first, then format, chart last
1. `create_spreadsheet` (optionally with named tabs; it cannot target a
folder - move it after with google-drive `google_drive_move_or_rename`).
@@ -35,9 +36,32 @@ data - call it first and keep the mapping. A new tab's sheetId is in
- `set_borders` (outer heavier than inner) or `add_banding` for
alternating row colors; `set_column_width` so nothing clips.
- `add_conditional_formatting` for thresholds worth seeing at a glance.
-4. `add_chart` LAST, after the data exists: first range is the domain
- (x axis / pie labels), later ranges are series; ranges live on the same
- tab as the chart. headerCount 1 uses first cells as labels.
+4. `add_chart` LAST, after the data exists.
+
+## Charts
+
+- In `add_chart`, the FIRST range is the domain (x axis / pie labels) and
+ later ranges are the series, one column each; anchoring a chart to the
+ wrong or reversed ranges is the most common chart bug. All ranges live on
+ the same tab as the chart. `headerCount: 1` uses the first cells as
+ series labels - include the header row in each range when you set it.
+- Place the chart over empty cells beside or below the table, not on top of
+ the data it plots.
+- A wrong chart is fixed in place, not worked around: `get_spreadsheet`
+ lists every tab's charts (chartId, title), `update_chart` replaces a
+ chart's spec keeping its position, `delete_chart` removes one. Never add
+ a second chart to paper over a bad first one.
+
+## Named and protected ranges
+
+- `add_named_range` makes formulas self-documenting (`=SUM(Budget)` instead
+ of `=SUM(B2:B13)`); write the formula with `parseInput: true`. Names
+ cannot look like cell references.
+- `protect_range` guards headers and formula cells (`warningOnly: true` by
+ default warns editors; `false` locks them). Protect AFTER the last write
+ to that range, or your own writes fight the protection.
+- `get_spreadsheet` lists both, with the ids `delete_named_range` and
+ `unprotect_range` need.
## Pitfalls
@@ -46,16 +70,17 @@ data - call it first and keep the mapping. A new tab's sheetId is in
- `merge_cells` keeps only the top-left value; write the text after merging.
- `sort_range` rewrites cell positions; sort before adding formulas that
reference the range, and exclude the header row from the sorted range.
-- `insert/delete_rows_or_columns` positions are 1-based (A=1); deletes are
- immediate and not undoable through the API.
+- `insert_rows_or_columns` / `delete_rows_or_columns` positions are 1-based
+ (A=1); deletes are immediate and not undoable through the API.
- `find_replace_sheet` is literal (never regex here) and can search inside
formulas with `searchFormulas: true`.
- Whole-tab reads can blow the byte cap; bound `cells` (e.g. "1:2000") on
tabs you have not sized and follow `nextCells` when truncated.
-## Verifying
+## Verify your work
-`read_range` (FORMATTED_VALUE shows what the user sees; UNFORMATTED_VALUE
-raw numbers) for values, `get_spreadsheet` for structure, and google-drive
-`google_drive_export_file` (pdf - renders every tab including charts and
-formatting) when the visual result matters.
+`read_range` for values (FORMATTED_VALUE shows what the user sees;
+UNFORMATTED_VALUE the raw numbers - use it to confirm numbers are numbers,
+not text). `get_spreadsheet` for structure: tabs, charts, named ranges,
+protections. google-drive `google_drive_export_file` (pdf renders every tab
+including charts and formatting) when the visual result matters.
diff --git a/third_party/google-slides/.cursor-plugin/plugin.json b/third_party/google-slides/.cursor-plugin/plugin.json
index 9036725a..b8c66328 100644
--- a/third_party/google-slides/.cursor-plugin/plugin.json
+++ b/third_party/google-slides/.cursor-plugin/plugin.json
@@ -1,7 +1,7 @@
{
"name": "google-slides",
"displayName": "Google Slides",
- "version": "1.0.0",
+ "version": "1.1.0",
"minClientVersions": {
"cursor": "3.22.0",
"grokbot": "0.30.0",
diff --git a/third_party/google-slides/CHANGELOG.md b/third_party/google-slides/CHANGELOG.md
index b3af0fbb..df2529a7 100644
--- a/third_party/google-slides/CHANGELOG.md
+++ b/third_party/google-slides/CHANGELOG.md
@@ -2,6 +2,10 @@
All notable changes to this plugin will be documented here.
+## 1.1.0
+
+- Refreshed the `google-slides` skill for the latest tool surface.
+
## 1.0.0 - initial release
- Logo: official Google Slides product icon on a padded white tile.
diff --git a/third_party/google-slides/skills/google-slides/SKILL.md b/third_party/google-slides/skills/google-slides/SKILL.md
index 50152666..34361980 100644
--- a/third_party/google-slides/skills/google-slides/SKILL.md
+++ b/third_party/google-slides/skills/google-slides/SKILL.md
@@ -1,77 +1,124 @@
---
name: google-slides
-description: Build and edit Google Slides through the google-slides MCP tools - slides, text boxes, shapes, images, tables, styling, and per-slide PNG rendering. Use when creating a deck, laying out slide elements, or verifying how a slide looks.
+description: Build and edit Google Slides through the google-slides MCP tools - slides, grids of cards and columns, text boxes, shapes, images, tables, styling, layout inspection, and per-slide PNG rendering. Use when creating a deck, laying out slide elements, matching a template, or verifying how a slide looks.
---
# Google Slides authoring
-## Handles and geometry
-
-Everything is addressed by objectId from `read_presentation`: slides, shapes,
-tables, images. Element positions and sizes are in POINTS from the slide's
-top-left; a standard slide is 720x405 pt. Plan a layout on that grid before
-inserting (e.g. title at y 40-60, content rows at y 150+, keep 40 pt margins)
-so elements do not overlap.
-
-## Sequencing a styled deck
-
-1. `create_presentation` (starts with ONE blank slide; it cannot target a
- folder - move it after with google-drive `google_drive_move_or_rename`).
- Delete the default placeholders when building a custom layout.
-2. `read_presentation` for the first slide's objectId, the layouts list,
- and the revisionId.
-3. Pick a PALETTE first: one background color, one card/accent color that
- harmonizes with it, one or two text colors. Every element uses it.
-4. Per slide: `set_slide_background`, then insert elements with their style
- INLINE - `insert_text_box` takes bold/fontSize/textColor/textAlignment,
- `insert_shape` additionally fillColor/outlineColor, so one call produces
- a finished element. `insert_image` aspect-fits into maxWidth/maxHeight
- and never crops; give it the box you want filled.
-5. More slides: `add_slide` with a predefined layout fills title/body
- placeholders in the same call; `move_slide` reorders.
-
-## Layout that reads well
-
-- Align to a grid: same y for elements in a row, equal sizes for peer cards,
- equal gaps (e.g. three 180 pt cards at x 60/270/480 on a 720 pt slide).
-- Whitespace is part of the design, but content should not huddle in one
- corner of an otherwise empty slide; spread rows vertically (title ~56,
- content ~180, footer ~330).
-- Tables are for genuinely tabular data on CONTENT slides; on title or
- summary slides prefer stat cards (`insert_shape` with text) or short
- bulleted placeholders.
-- `set_text_style` restyles ALL text of one shape (or a table cell via
- cellRow/cellColumn); it does not do partial ranges.
-
-Pass `requiredRevisionId` from the latest read (or the previous write's
-response) so concurrent edits fail the write instead of corrupting layout.
-
-## Tables
-
-`insert_table` makes an empty grid; fill cells with `set_shape_text` using
-`pageObjectId` = the slide, `objectId` = the table, and
-`cell: {rowIndex, columnIndex}` (0-based, EXACTLY those key names). Note
-`set_shape_text` REPLACES the whole cell/shape text.
-
-## Placeholders vs free elements
-
-`add_slide` with layout TITLE_AND_BODY etc. fills placeholders - the fastest
-way to a clean, theme-consistent slide. Free `insert_text_box`/`insert_shape`
-give exact control but inherit no theme styling; set fonts and colors
-yourself with `set_text_style`.
-
-## Position changes
-
-`update_element_position` moves an element to an absolute point position; it
-does not resize (the API resizes via transform scale against intrinsic size,
-which is unreliable across element kinds - delete and re-create at the right
-size instead).
-
-## Verifying visually
-
-`get_slide_thumbnail` renders ONE slide as PNG (base64; LARGE = 1600 px wide)
-- the ground truth for layout checks; read_presentation cannot tell you
-whether elements overlap. For the whole deck use google-drive
-`google_drive_export_file` (pdf). Speaker notes: write with `set_shape_text`
-using pageObjectId = the slide's notesPageId and objectId =
-speakerNotesObjectId from `read_presentation`.
+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 `card`s, 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).