fix: the slide prompt carries the rules the renderer actually enforces

Generated a deck with ds-deepseek-v4-flash and rendered it to look at. The model
produced exactly the six headings it was asked for; the deck came out with eight
slides. The extra ones were pandoc's, not the model's.

Two rules, both found by rendering rather than reading:

  - pandoc splits a slide after a table. Anything following one becomes a new
    slide with no title — that was the stray "Key differentials to consider:"
    slide floating with no heading.
  - A table with no blank line before it is not parsed as a table at all. It
    renders as literal pipe characters in the preceding paragraph.

And one that was visible on the slide itself: a nested ordered list inside a
bullet ran off the bottom.

None of these are the model failing. A cheap model writes perfectly good slide
markdown — bold, italics, nested lists and a table with a subscript all came
through correctly. It just needs to be told the shape the renderer wants.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU

src/routes/learningAI.js

@@ -167,6 +167,13 @@ Then each slide separated by ---. Guidelines:
 - First slide: title slide with presentation name and brief subtitle
 - Use # for slide titles
 - Use bullet points (- ) for lists, keep them concise (max 5 bullets per slide)
+- A slide that contains a table must contain ONLY that table and its heading.
+  Anything after a table starts a new, untitled slide when the deck is built.
+- Leave a blank line before and after every table, or it is not read as a table
+  at all and appears as literal pipe characters on the slide.
+- Do not nest lists more than one level deep, and do not put an ordered list
+  inside a bullet: it overfills the slide.
+- Prefer more slides with less on each. A slide should hold one idea.
 - Include a summary/key takeaways slide at the end
 - Do NOT include HTML tags or inline styles`;
     }

test/backend-hardening.test.js

@@ -330,3 +330,17 @@ test('a deck can only embed images the requester owns', () => {
   assert.match(src, /workdir = await fsp\.mkdtemp\(/);
   assert.match(src, /\} finally \{[\s\S]{0,200}rm\(workdir, \{ recursive: true, force: true \}\)/);
 });
+
+test('the slide prompt carries the rules pandoc actually enforces', () => {
+  // Found by generating a deck with ds-deepseek-v4-flash and rendering it: the
+  // model produced exactly the 6 headings asked for, but the deck came out with
+  // 8 slides. pandoc splits a slide after a table, and the remainder becomes an
+  // untitled orphan. A table with no blank line before it is not parsed as a
+  // table at all — it renders as literal pipe characters.
+  const src = read('src/routes/learningAI.js');
+  assert.match(src, /A slide that contains a table must contain ONLY that table/);
+  assert.match(src, /Leave a blank line before and after every table/);
+  // And the overfull slide in that same test: a nested ordered list inside a
+  // bullet ran past the bottom of the slide.
+  assert.match(src, /do not put an ordered list\s*\n\s*inside a bullet/);
+});