Google Slides

Google Slides (Template-Driven Deck Editor)

Generate a Google Slides deck by picking pattern slides from a template deck, duplicating them, filling content via style-preserving text replacement, and hiding the originals. Every visible slide is a copy of a pre-styled pattern, so the template's fonts, colors, layouts, and theme survive without any explicit styling work from the agent.

Why this shape? The Google Slides API lets you edit text and structure but gives you no judgment about typography or layout. Treating each template slide as one finished pattern (not a parts catalog) keeps the agent out of styling decisions. replaceAllText edits text runs in place, so run-level styling — font, size, weight, color — is preserved automatically.

Prerequisites

Both connectors must be connected in Treasure Studio → Settings → Connected Services:

Confirm both are connected before proceeding. If either is missing, stop and ask the user to connect it.

Template Requirements (Author Contract)

The template deck must follow the complete-slide pattern convention:

• Each slide is one reusable layout, ready to be used whole
• No "parts catalog" slides that group multiple tables, CTAs, or boxes for users to extract pieces from — the Slides API cannot move elements between slides without losing fidelity
• Placeholder text uses recognizable tokens like [Title], [Body], [Metric Name], [CTA] — the agent replaces them by exact string match
• Slide titles, when present, follow Pattern: <layout description> (e.g., Pattern: Stat Callout (3 columns)) so the agent can pick from the catalog confidently

If the template violates these rules, explain the constraint and ask the user to update the template — working around it silently produces decks that do not match the template's design language.

Tools You Will Use

All tools are in the Studio MCP server (mcp__tdx-studio__*). Full tool schemas live in the tdx repo at studio/electron/services/google-tools.ts under createGoogleSlidesTools.

google_slides_replace_text and google_slides_batch_update take YAML-encoded lists — see references/batch-update-recipes.yaml for the recipe library. Copy a recipe, fill in object IDs, pass the YAML string straight through as requests_yaml or replacements_yaml.

Workflow

The full step-by-step workflow lives in references/workflow.md. Read it before planning a deck — it explains what to verify at each step and how to recover from common mistakes. Short version:

1. Confirm both connectors, collect template URL + brief
2. Copy template deck        (google_drive_copy)
3. List slides               (google_slides_list_slides)
4. Plan deck                 (propose slide plan, get user sign-off)
5. Inspect chosen patterns   (google_slides_get_slide per plan entry;
                               inventory text, table cells, empty placeholders,
                               AND images — template stock images are replaceable content)
6. Duplicate into final position (google_slides_duplicate_slide with insertion_index —
                                  either reverse-iterate at index 0, or forward-iterate
                                  with growing index; see workflow.md Step 6)
7. Fill content              (google_slides_replace_text for text, google_slides_batch_update insertText for empty
                               placeholders, google_slides_batch_update replaceImage for images. Agent-
                               generated images go through generate_image → google_drive_upload
                               → google_drive_share → replaceImage)
8. Hide all used originals   (one google_slides_hide_slides call with the full array)
9. QA                        (load google-slides-review skill — independent
                               sub-agent review; loop until 0 findings)
10. Return the working deck URL plus a summary of patterns used

Prefer hiding a pattern slide (step 8) over deleting it: hiding is reversible, keeps the template readable as a reference, and lets you recover the pattern if QA surfaces a problem.

Pattern Selection

How to read thumbnails, match a brief to patterns, and avoid misclassification is covered in references/pattern-selection.md. Read it when building the slide plan in step 4.

QA

Step 9 hands off to the google-slides-review skill — see that skill's Inputs and Loop contract for how to invoke and when to stop. The leak sub-agent needs the per-element text snapshot captured during Step 5; do not skip that capture.

Failure Modes and Recovery

Non-obvious failures the agent is prone to. Generic issues (rate limits, YAML syntax errors) are not listed — the agent handles those from its own reading of error messages.