feat(mcp): add resources and prompts

Resources: pascal://scene/current (JSON), /scene/current/summary
(markdown with per-level counts, floor areas, bbox), /catalog/items
(returns catalog_unavailable in headless mode), and the templated
pascal://constraints/{levelId} which exposes slabs + wall footprints
via @pascal-app/core/wall helpers.

Prompts: from_brief (generate scene from a natural-language brief),
iterate_on_feedback (minimal-diff patch proposals), and
renovation_from_photos (orchestrates the vision tools).

17 tests, all passing.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

packages/mcp/src/prompts/iterate-on-feedback.ts

@@ -0,0 +1,47 @@
+import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js'
+import { z } from 'zod'
+import type { SceneBridge } from '../bridge/scene-bridge'
+
+const PREAMBLE = [
+  'You are iterating on an existing Pascal scene based on user feedback.',
+  'Given the current state (read via the `pascal://scene/current` resource) and the user feedback below, propose the MINIMUM set of `apply_patch` operations that satisfies the feedback.',
+  'Rules:',
+  '  - Prefer updates over create+delete pairs when a field change will do.',
+  '  - Do not re-create nodes that already exist.',
+  '  - Do not touch nodes that are unrelated to the feedback.',
+  '  - Bundle related mutations into a single `apply_patch` call so they share one undo step.',
+  '  - Respond ONLY with tool calls. No prose.',
+].join('\n')
+
+/**
+ * Build the user-facing prompt text for `iterate_on_feedback`.
+ * Pure function for testability.
+ */
+export function buildIterateOnFeedbackPrompt(args: { feedback: string }): string {
+  return [PREAMBLE, '', '## User feedback', args.feedback.trim()].join('\n')
+}
+
+export function registerIterateOnFeedback(server: McpServer, _bridge: SceneBridge): void {
+  server.registerPrompt(
+    'iterate_on_feedback',
+    {
+      title: 'Iterate on a scene from user feedback',
+      description:
+        'Produces a minimal-diff plan of apply_patch calls in response to user feedback on the current scene.',
+      argsSchema: {
+        feedback: z.string(),
+      },
+    },
+    async ({ feedback }) => ({
+      messages: [
+        {
+          role: 'user',
+          content: {
+            type: 'text',
+            text: buildIterateOnFeedbackPrompt({ feedback }),
+          },
+        },
+      ],
+    }),
+  )
+}