feat(take-notes): add Google Slides as a source

[davertor]

What and why

/take-notes <presentation URL> now handles a Google Slides deck, and brings its
diagrams into the note.

Google exports any link-visible deck as .pptx without an API key, and a pptx is
a zip of XML — so scripts/slides.py stays stdlib only, per the house rule.

The part that took the work: a deck's diagrams are frequently the entire
argument, and most of them exist in no image file. A pipeline drawn as boxes
and arrows is native Slides shapes, so extracting ppt/media/ would return the
pasted screenshots and silently miss every drawn diagram. Rendering the whole
slide catches both — and the slide's Google page id, which export/png?pageid=
needs, turns out to be recoverable only from the shape names on its notes page
(Google Shape;117;<id>:notes). That is the one thing the pptx export gives that
nothing else does.

slides.py also reports per slide how many images, shapes and connectors it
carries. connectors is the strongest signal for a figure worth keeping: boxes
alone are a layout, boxes joined by arrows are an explanation.

Speaker notes come back alongside the slide text, and references/slides.md makes
reading them a requirement rather than a tip — the bullets are what the audience
sees, the notes are what the presenter was going to say. The header states whether
any exist (on 12 of 16 slides / none written on any slide) so a deck nobody
annotated cannot be mistaken for notes that failed to parse.

Two smaller changes ride along:

• SKILL.md's Article figures rule becomes Source figures, covering articles,
  papers and decks under one cap of 3.
• references/web.md's figure check lacked -L, so any redirecting image host
  (a CDN, or Google's own export) read as a failure and a good figure got dropped.

Verified end to end

Ran the whole flow against a real 16-slide deck: three figures embedded, each
confirmed 200 image/png. Also checked the raw XML of that deck's 16 notes pages
to confirm its empty speaker notes were genuinely empty rather than a parse miss —
which is what prompted the header line.

slides.py is new to the self-check loop, so it is added to CI, CONTRIBUTING.md,
and both issue/PR templates (which had already drifted apart on retag).

Checks

• for s in render notes gallery export transcript tags retag slides; do uv run skills/take-notes/scripts/$s.py --selftest; done passes
• No new third-party dependency (see CONTRIBUTING.md)
• English throughout — code, comments, commit messages

If you touched…

• a note template — untouched; notes.py --selftest still passes
• SKILL.md Sections — the <li><strong>term</strong> — definition</li> shape is unchanged, so Anki export still finds cards
• disable-model-invocation — unchanged, so agents/openai.yaml needs no counterpart
• the version — 1.1.0 in SKILL.md and all three manifests, with a CHANGELOG.md entry
• anything user-facing — README.md updated (no README.es.md in the tree)

Known trade-off

Figures hotlink Google's export endpoint rather than being downloaded and inlined.
The deck has to be publicly fetchable for acquisition to work at all, so this costs
nothing today, and it keeps the Markdown export clean. It does mean a figure renders
the deck's current slide and breaks if sharing is later tightened —
references/slides.md says so. Inlining as data URIs is the upgrade if that starts
to matter.

🤖 Generated with Claude Code