Add the design-diff demo: recipe assistant vs recipe + nutrition

A genuine variant (recipe app plus nutrition and health guidance at 800k
MAU), generated through the real pipeline, and the agentify diff report
comparing it to the original. The diff shows the sharp contrast: the
weight class rises lightweight to enterprise while the escalation rung
stays at 1, driven by health-grade domain harm rather than tenancy, and
the skill refuses individualized medical advice.

- examples/recipe-nutrition.* (variant design + architecture) and
  recipe-nutrition.diff.html (the comparison report)
- build-site.mjs also publishes *.diff.html; the diff is in the site
  gallery and linked from the README
- eval: recipe-nutrition seeded (rung 1 + enterprise weight class), all
  6 seeded cases pass

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: avnath13

README.md

@@ -184,6 +184,8 @@ Three complete generated designs are committed in [`examples/`](examples/); open
 
 Each design embeds diagrams drawn with the agent-native vocabulary: a component architecture plus a request sequence, a ticket lifecycle state machine, or a voice pipeline.
 
+And a [**design comparison**](https://avnath13.github.io/agentify/examples/recipe-nutrition.diff.html) produced by `agentify diff`: the recipe assistant with health guidance added and 800k users, showing what the change did (the weight class rises to enterprise while the rung stays the same, and the skill refuses to give individualized medical advice).
+
 <picture>
   <source media="(prefers-color-scheme: dark)" srcset="docs/assets/drive-through-voice-dark.png">
   <img alt="Architecture for a drive-through voice agent: a store-edge boundary with audio edge, streaming ASR and streaming TTS; an order-agent runtime with input and output guardrails, a model gateway with a cached menu prefix, order state, a menu store, a bounded order agent, barge-in, POS submit as a tier-2 action, staff takeover for allergens and low confidence, and a voice eval gate" src="docs/assets/drive-through-voice-light.png" width="100%">

docs/index.html

@@ -274,6 +274,12 @@ <h3>Drive-through voice agent</h3>
         <p>A cascaded voice design with the conversational latency budget, turn-taking (VAD, endpointing, barge-in), and audio-native evaluation, drawn with the speech component types.</p>
         <span class="open">Open design &rarr;</span>
       </a>
+      <a class="g-card" href="examples/recipe-nutrition.diff.html">
+        <span class="kind">agentify diff</span>
+        <h3>Compare: add nutrition and scale</h3>
+        <p>The recipe assistant with health guidance added and 800k users. A visual diff shows what the change did: the weight class jumps to enterprise while the rung stays the same, and the skill refuses to give individualized medical advice.</p>
+        <span class="open">Open comparison &rarr;</span>
+      </a>
     </div>
   </div>
 </section>

eval/cases.jsonl

@@ -10,3 +10,4 @@
 {"id":"drive-through-voice","prompt":"A voice agent that takes drive-through food orders.","answers":"Quick-service restaurant chain, noisy environment, orders go to the POS, human staff nearby, weekly regional menus, order accuracy is the top metric.","design":"examples/drive-through-voice.design.html","expect":{"mustMention":["latency","cascaded","barge-in","endpoint"],"mustNot":[]}}
 {"id":"insurance-photo-claims","prompt":"An agent that assesses vehicle damage from a photo a customer uploads and drafts a claim.","answers":"Auto insurer, customer uploads photos, a human adjuster approves payouts.","expect":{"mustMention":["multimodal","grounding","hallucination","human"],"mustNot":[]}}
 {"id":"lead-triage","prompt":"An agent that triages inbound sales leads: enrich from CRM, score, and draft a reply for a rep to approve.","answers":"300 leads/day, CRM has the data, reps approve every send.","expect":{"rung":3,"mustMention":["human","approv","enrich"],"mustNot":[]}}
+{"id":"recipe-nutrition","prompt":"The recipe app, but it also gives basic nutrition and dietary-health guidance, at 800k MAU.","answers":"Same catalog and team, 800k MAU, wants calories/macros and answers about diabetes/hypertension, nervous about wrong dietary-health advice.","design":"examples/recipe-nutrition.design.html","expect":{"rung":1,"weightClass":"enterprise","mustMention":["deterministic","disclaimer","fail-closed"],"mustNot":[]}}