Thinkcell

think-cell automation

A skill for generating think-cell charts and elements correctly, in the right format, following a company template.

Core principle: think-cell is template-driven

think-cell is not a charting library Claude can call directly. It is a PowerPoint COM add-in. To produce a think-cell chart programmatically, the workflow is always:

1. A PowerPoint template exists (.pptx) containing think-cell elements (chart, table, text field, Harvey ball, etc.) that each have a name assigned via the think-cell mini toolbar (AddRangeData Name).
2. Data is supplied separately — as JSON (.ppttc file) or as an Excel workbook range.
3. think-cell merges them when the .ppttc is opened, when PresentationFromTemplate runs, or when UpdateBatch.Send() runs.

This means the formatting (fonts, colors, layout, label positions, source line, page numbers) lives in the template, not in the code Claude writes. Claude's job is to (a) produce the correct data payload and (b) wire it to the correctly-named element. Everything else — house style — comes from the template, which is why charts produced this way automatically match the rest of the deck.

Therefore, before doing anything, Claude must know:

• Which template is being used (path to the .pptx)
• Which named elements are in it (and what kind: chart? text field? table?)
• For each chart element, its datasheet layout (default? 100%= row? Size column for Mekko? etc.)

If any of these are unknown, ask the user, or — if a template file is available — inspect it before generating code.

French ↔ English glossary (the user may speak either)

think-cell's UI exists in French. Recognise both labels — and when the user writes in French, respond in French and use the French chart/element names in explanations (but keep code identifiers — JSON keys, VBA function names — in English; those are fixed by the API).

Common French request patterns and what they mean:

• "Fais un graphique en cascade avec ces chiffres" → waterfall via JSON automation (.ppttc).
• "Mets ce tableau dans le modèle de la boîte" → fill named element via JSON or UpdateBatch.
• "Comme ailleurs dans la présentation" / "Respecter la charte" → route through the corporate .pptx template; don't try to set fonts/colors from code (see references/style-and-branding.md).
• "La feuille de données doit avoir la ligne 100 %=" → enable the 100%= row in the template's datasheet layout.

Decision tree: which path to use

Default recommendation: If unsure, use JSON automation (.ppttc). It's the cleanest path, requires no COM setup, works without Claude having to script Office, and produces a .pptx the user can immediately edit. The Excel COM path is for recurring report-refresh workflows.

Required reading before generating output

Before generating any think-cell artifact (.ppttc, VBA macro, C# code, Python COM script, template instructions), Claude must read the relevant reference file(s):

• Always read references/json-automation.md if producing a .ppttc file. Get the exact cell types right (string, number, date, percentage, fill, null); date format is YYYY-MM-DD; percentages are bare numbers without %; decimal separator is always .; the first cell of row 1 is always null (or anything — it's discarded).
• Always read references/chart-datasheets.md for the specific chart type being produced, to know which rows mean categories vs. series vs. 100%= totals vs. Mekko widths vs. Gantt anchor dates.
• Always read references/excel-automation.md if generating VBA, C#, or Python COM code. Get the Transposed argument right — it differs per element type.
• Read references/templates-and-naming.md if the user needs help structuring or inspecting a template. When a .pptx template is uploaded, run scripts/inspect_template.py on it first — it lists slides, shapes, and any think-cell custom XML found, and gives you a name-candidate list to confirm with the user.
• Read references/style-and-branding.md if the user asks about matching company colors/fonts, datasheet fill colors, or style files.

Reading the reference is mandatory, not optional. The schema details (especially Mekko's Size column, scatter/bubble's dimension layout, and Gantt's anchor-date convention) are easy to get wrong from memory.

The naming contract

Every think-cell element in the template that Claude wants to fill must have an AddRangeData Name (or AddRangeImage Name for table images) assigned via the mini toolbar in PowerPoint. Names are case-insensitive and must be unique within the template. If two elements share a name, think-cell fills them with the same data.

When proposing names to the user, use a clear convention:

• SlideTitle, Chapter1Title — for automation text fields holding titles
• RevenueChart, MarginByRegion — for charts
• KPITable, PnLTable — for tables
• Source — for the source footnote text field (if templated)

If the user hasn't set names yet, the first deliverable is a checklist of what to name in the template (see references/templates-and-naming.md).

When generating .ppttc files

1. Read references/json-automation.md first.
2. Read references/chart-datasheets.md for the chart types involved.
3. Use the helper at scripts/build_ppttc.py to construct the JSON — it gets cell types right by construction.
4. Validate with scripts/validate_ppttc.py before presenting the file to the user.
5. Save the output as something.ppttc (NOT .json) so think-cell's file association picks it up on double-click. The file content is JSON; only the extension differs.
6. Tell the user the file pairs with a specific template (state which .pptx) and what each named element receives.

When generating VBA / C# / Python COM code

1. Read references/excel-automation.md first.
2. Always show the user the prerequisites: think-cell installed, COM add-in thinkcell.addin loaded, PowerPoint Object Library referenced (VBA) or interop assemblies added (C#).
3. Default to UpdateBatch (modern API) over UpdateChart (deprecated).
4. Get Transposed right — re-check the table in the reference. The default for most charts is rows=series, columns=categories, so if the Excel layout puts series in columns, set Transposed = True.
5. Wrap COM objects with explicit close / quit / release to avoid orphaned Office processes (especially for Python win32com).

When the user's request is vague

Ask clarifying questions in this priority order:

1. Which template / which style? (path to .pptx, or "no template yet")
2. What chart type(s)? (waterfall, stacked column, Mekko, Gantt, …)
3. Where is the data? (inline in the request, Excel file, JSON, database)
4. Refresh cadence? (one-off → JSON automation; recurring report → Excel automation)

If the user provides data inline and doesn't mention a template, proceed with JSON automation against a minimal default template and call this out explicitly: "I'll generate a .ppttc that targets a default template — for this to render in your house style, point it at your corporate template instead by changing the template field."

Adapting to the company template (the "elsewhere in the template" requirement)

The skill cannot read minds about house style. It can, however:

• Preserve house style automatically by routing data through the template (since fonts/colors live there).
• Inspect an uploaded .pptx template to enumerate named elements, chart types, and datasheet layouts before generating data — do this whenever a template is provided.
• Match Excel formatting to chart features via the "Use Datasheet Fill" option (see references/style-and-branding.md) so that per-cell fill colors flow into the chart.

If the user uploads a template, the first step is always: open it, list its named elements with their types and datasheet layouts, and confirm with the user before generating data.

Output conventions

• .ppttc files: pretty-printed JSON, 2-space indent, trailing newline.
• VBA: Option Explicit at top, explicit Dim declarations, comment the prerequisites block at top.
• C#: full using block, Main() that opens/closes Office cleanly.
• Python (win32com): wrap in try/finally with explicit Quit() calls, and use pythoncom.CoInitialize() if running outside the main thread.

What this skill does NOT do

• It does not install think-cell. The user must have a licensed installation.
• It does not work without an AddRangeData Name on each target element — naming is a manual step in the template.
• It does not work on macOS for the COM/Excel automation path (Windows only). JSON automation works on macOS provided think-cell for Mac is installed.
• It does not modify the visual style of a chart programmatically beyond per-cell datasheet fill colors. Style changes belong in the template or in a think-cell style file (.style).