ingest-slide-templates

name

ingest-slide-templates

description

Ingest one or more slides from a PowerPoint deck into the pptx-gen template library and document each one. Use when the user gives a source .pptx file (often a single slide or a deck downloaded from Google Slides) plus a list of slide numbers and a template name, and asks to import, ingest, or add slides as reusable templates in templates/. Runs the ingest CLI for each slide, then runs the draft-slide-template-description workflow on each imported template so it ships with a real description instead of a TODO stub.

Ingest slide templates

Purpose

Turn slides from a real deck into reusable templates in templates/, fully documented and ready to pick when building a deck.

This skill chains two steps:

1. Ingest each requested slide with the CLI (ingest --slide). This clones the slide, extracts its text fields and fonts, renders a screenshot (if LibreOffice is installed), and writes a description.md TODO stub.
2. Describe each imported template by running the draft-slide-template-description workflow, which replaces that stub with a short, accurate description.

The output is one finished template folder per slide, with template.pptx, template.yml, fields.yml, an optional screenshot, and a real description.md.

Fixed paths

• Ingest CLI: run from the repo root with npm run cli -- ingest ...
• Template library: templates/<template-name>/

Use Node 20 or newer.

Required inputs (stop if missing)

You need all three before doing anything. If any is missing, stop and ask.

1. Source .pptx file path. A local file. For Google Slides, the user exports it first (File > Download > Microsoft PowerPoint .pptx) and gives you the path. We do not download from cloud drives inside this skill: real decks are large, and a manual export is reliable.
2. Slide numbers to ingest. A list, for example 3, 5, 16, or a single one.
3. Template name(s). What each imported slide should be called.
   • One slide: use the name as given (for example two-column-comparison).
   • Several slides: ask for one name per slide. If the user gives a single base name, append -slide-NN per slide (zero-padded), matching --split.

Steps

1. Confirm inputs and inspect the source

Check the file exists and count how many slides it really has:

unzip -l "<source.pptx>" | grep -cE 'ppt/slides/slide[0-9]+\.xml'

Reconcile the numbers the user gave against this count:

• One slide in the file. Common: the user exported just the slide they care about, so the file holds only it even though they call it "slide 16". Ingest it as position 1, and say so.
• Many slides, all requested numbers in range. Treat the numbers as positions and ingest each.
• A requested number is out of range and the file has more than one slide. Do not guess. Stop and ask which position they mean.

2. Ingest each slide

For each (position, template name) pair, run from the repo root. First confirm the target name is free (ls templates/<name>); if it exists, ask before overwriting.

npm run cli -- ingest \
  --source "<source.pptx>" \
  --template "<template-name>" \
  --slide <position>

The CLI ingests one slide per call, so loop it once per slide. After each ingest, check the folder was created. If a screenshot-warnings.md appeared, read it and tell the user (a missing screenshot just means LibreOffice is not installed; the template is still valid).

To split every slide of a multi-slide deck in one call, use --split instead of --slide.

3. Describe each imported template

For every template you just imported, run the draft-slide-template-description workflow on it. Do each as its own job; do not blend observations across slides.

4. Report

Tell the user, per slide: the template name and folder path, the field and font counts from the ingestion report, a one-line summary of the description you wrote, and any screenshot warnings.

Notes and traps

• The slide number is not always the file position. A single-slide export is the normal case when the user grabs one slide from a big deck.
• Do not invent layout. The screenshot (or, if absent, the fields) is the source of truth for the description, not the slide name.
• Template names are folder names. Keep them kebab-case and descriptive of intent (two-tier-comparison-cards), not of the source deck.