diff --git a/mateclaw-server/src/main/java/vip/mate/wiki/dto/ImageRef.java b/mateclaw-server/src/main/java/vip/mate/wiki/dto/ImageRef.java
new file mode 100644
index 00000000..340b64c2
--- /dev/null
+++ b/mateclaw-server/src/main/java/vip/mate/wiki/dto/ImageRef.java
@@ -0,0 +1,20 @@
+package vip.mate.wiki.dto;
+
+/**
+ * One image reference parsed out of a markdown body.
+ *
+ * <p>Maps to a single {@code ![alt](url)} occurrence. Search results carry
+ * a list of these so the UI can render thumbnails alongside the page hit
+ * without re-scanning the markdown on the client.
+ *
+ * @param fullMatch verbatim {@code ![alt](url)} as it appeared in the source
+ * @param alt       alt text (may be empty if the markdown wrote {@code ![](url)})
+ * @param url       resource URL — local path, absolute http(s) URL, or data URI
+ *
+ * @author MateClaw Team
+ */
+public record ImageRef(
+        String fullMatch,
+        String alt,
+        String url
+) {}
diff --git a/mateclaw-server/src/main/java/vip/mate/wiki/dto/PageSearchResult.java b/mateclaw-server/src/main/java/vip/mate/wiki/dto/PageSearchResult.java
index d0e458ae..cf1b8395 100644
--- a/mateclaw-server/src/main/java/vip/mate/wiki/dto/PageSearchResult.java
+++ b/mateclaw-server/src/main/java/vip/mate/wiki/dto/PageSearchResult.java
@@ -3,7 +3,15 @@ package vip.mate.wiki.dto;
 import java.util.List;
 
 /**
- * RFC-032: Enhanced search result with snippet, match metadata, and relevance reason.
+ * Enhanced search hit returned by hybrid retrieval.
+ *
+ * <p>Carries the matched page identity ({@link #slug}, {@link #title},
+ * {@link #summary}), a query-specific {@link #snippet} extracted from the
+ * page body, the per-mode score breakdown ({@link #matchedBy}) and a
+ * human-readable {@link #reason} for why the page surfaced. The
+ * {@link #imageRefs} field is populated with up to N image references
+ * pulled from the page body so the UI can render thumbnails inline with
+ * the hit list.
  */
 public record PageSearchResult(
     String slug,
@@ -12,5 +20,13 @@ public record PageSearchResult(
     String snippet,
     List<String> matchedBy,
     String reason,
-    double score
-) {}
+    double score,
+    List<ImageRef> imageRefs
+) {
+
+    /** Convenience factory used by callers that have not (yet) computed image refs. */
+    public static PageSearchResult of(String slug, String title, String summary, String snippet,
+                                       List<String> matchedBy, String reason, double score) {
+        return new PageSearchResult(slug, title, summary, snippet, matchedBy, reason, score, List.of());
+    }
+}
diff --git a/mateclaw-server/src/main/java/vip/mate/wiki/service/HybridRetriever.java b/mateclaw-server/src/main/java/vip/mate/wiki/service/HybridRetriever.java
index 3be05b6e..e63cf404 100644
--- a/mateclaw-server/src/main/java/vip/mate/wiki/service/HybridRetriever.java
+++ b/mateclaw-server/src/main/java/vip/mate/wiki/service/HybridRetriever.java
@@ -5,6 +5,7 @@ import lombok.extern.slf4j.Slf4j;
 import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.stereotype.Service;
 import vip.mate.wiki.WikiProperties;
+import vip.mate.wiki.dto.ImageRef;
 import vip.mate.wiki.dto.PageSearchResult;
 import vip.mate.wiki.dto.RelatedPageResult;
 import vip.mate.wiki.dto.WikiPageLite;
@@ -35,24 +36,31 @@ public class HybridRetriever {
     private final WikiProperties properties;
     private final WikiPageMapper pageMapper;
     private final WikiMetrics metrics;
+    private final WikiContentNormalizer contentNormalizer;
 
     @Autowired(required = false)
     private WikiRelationService relationService;
 
     private static final double RELATION_BOOST = 0.15;
 
+    /** Cap how many image references one search hit carries — large pages can have
+     *  dozens; we only need a couple of thumbnails for the result panel. */
+    private static final int MAX_IMAGE_REFS_PER_HIT = 3;
+
     public HybridRetriever(WikiPageService pageService,
                             WikiChunkService chunkService,
                             WikiEmbeddingService embeddingService,
                             WikiProperties properties,
                             WikiPageMapper pageMapper,
-                            WikiMetrics metrics) {
+                            WikiMetrics metrics,
+                            WikiContentNormalizer contentNormalizer) {
         this.pageService = pageService;
         this.chunkService = chunkService;
         this.embeddingService = embeddingService;
         this.properties = properties;
         this.pageMapper = pageMapper;
         this.metrics = metrics;
+        this.contentNormalizer = contentNormalizer;
     }
 
     public enum Mode { KEYWORD, SEMANTIC, HYBRID }
@@ -132,10 +140,17 @@ public class HybridRetriever {
             if (lite.isSystem()) continue;
 
             String snippet = null;
+            List<ImageRef> imageRefs = List.of();
             if (!ri.matchedBy.contains("relation_boost")) {
                 String content = pageMapper.selectContentById(ri.pageId);
                 if (content != null) {
                     snippet = SnippetExtractor.extract(content, query);
+                    // Surface up to N image references inline with the hit so the UI
+                    // can render thumbnails without re-fetching the page body.
+                    List<ImageRef> all = contentNormalizer.extractImageRefs(content);
+                    imageRefs = all.size() > MAX_IMAGE_REFS_PER_HIT
+                            ? all.subList(0, MAX_IMAGE_REFS_PER_HIT)
+                            : all;
                 }
             }
 
@@ -151,7 +166,7 @@ public class HybridRetriever {
             results.add(new PageSearchResult(
                 lite.slug(), lite.title(), lite.summary(),
                 snippet != null ? snippet : lite.summary(),
-                ri.matchedBy, reason, ri.score));
+                ri.matchedBy, reason, ri.score, imageRefs));
         }
         metrics.recordRetrieval(mode.name().toLowerCase(),
                 Duration.ofNanos(System.nanoTime() - startNanos), results.size());
diff --git a/mateclaw-server/src/main/java/vip/mate/wiki/service/WikiContentNormalizer.java b/mateclaw-server/src/main/java/vip/mate/wiki/service/WikiContentNormalizer.java
index 7b9275bd..010cfb30 100644
--- a/mateclaw-server/src/main/java/vip/mate/wiki/service/WikiContentNormalizer.java
+++ b/mateclaw-server/src/main/java/vip/mate/wiki/service/WikiContentNormalizer.java
@@ -6,6 +6,14 @@ import org.jsoup.nodes.Document;
 import org.jsoup.nodes.Element;
 import org.jsoup.select.Elements;
 import org.springframework.stereotype.Component;
+import vip.mate.wiki.dto.ImageRef;
+
+import java.util.ArrayList;
+import java.util.HashSet;
+import java.util.List;
+import java.util.Set;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
 
 /**
  * RFC-051 PR-1c: source-type-specific text normalization.
@@ -101,4 +109,34 @@ public class WikiContentNormalizer {
         String unified = text.replace("\r\n", "\n").replace('\r', '\n');
         return unified.replaceAll("\n{3,}", "\n\n");
     }
+
+    /** Markdown {@code ![alt](url)} pattern. {@code url} captures up to the first
+     *  whitespace or closing paren so titles like {@code ![alt](url "title")} stop
+     *  at the URL boundary; the title segment is intentionally discarded. */
+    private static final Pattern MD_IMAGE_REF = Pattern.compile("!\\[([^\\]]*)]\\(([^)\\s]+)\\)");
+
+    /**
+     * Extracts every {@code ![alt](url)} occurrence in the input as an
+     * {@link ImageRef}. URLs are deduplicated — the first occurrence wins
+     * so the alt text closest to the document start is preserved when the
+     * same image appears multiple times.
+     *
+     * @param markdown raw markdown body; null/blank input returns an empty list
+     * @return ordered, URL-deduplicated list of references, never null
+     */
+    public List<ImageRef> extractImageRefs(String markdown) {
+        if (markdown == null || markdown.isBlank()) {
+            return List.of();
+        }
+        List<ImageRef> refs = new ArrayList<>();
+        Set<String> seenUrls = new HashSet<>();
+        Matcher m = MD_IMAGE_REF.matcher(markdown);
+        while (m.find()) {
+            String url = m.group(2);
+            if (url == null || url.isBlank()) continue;
+            if (!seenUrls.add(url)) continue;
+            refs.add(new ImageRef(m.group(0), m.group(1), url));
+        }
+        return refs;
+    }
 }
diff --git a/mateclaw-ui/src/views/Wiki/components/ImageLightbox.vue b/mateclaw-ui/src/views/Wiki/components/ImageLightbox.vue
new file mode 100644
index 00000000..e096e0ca
--- /dev/null
+++ b/mateclaw-ui/src/views/Wiki/components/ImageLightbox.vue
@@ -0,0 +1,71 @@
+<template>
+  <el-image-viewer
+    v-if="open"
+    :url-list="urls"
+    :initial-index="activeIndex"
+    :hide-on-click-modal="true"
+    :teleported="true"
+    @close="close"
+  />
+</template>
+
+<script setup lang="ts">
+/**
+ * Lightbox overlay for inline wiki page images.
+ *
+ * <p>The component is mounted once per wiki page view and exposes a
+ * single imperative entry point — {@link attach} — that the parent
+ * calls after a markdown render lands. attach() walks every <img>
+ * inside the supplied container, attaches a click handler that opens
+ * the lightbox at the image's index, applies the cursor cue, and
+ * marks the element as bound so subsequent attach() calls are
+ * idempotent.
+ *
+ * <p>Element Plus ships {@code ElImageViewer} as a teleported
+ * fullscreen overlay; the heavyweight scroll/zoom UI comes from
+ * upstream so this component stays a thin coordinator.
+ */
+import { ref } from 'vue'
+import { ElImageViewer } from 'element-plus'
+
+const open = ref(false)
+const urls = ref<string[]>([])
+const activeIndex = ref(0)
+
+const BOUND_FLAG = 'wikiLightboxBound'
+
+function close() {
+  open.value = false
+}
+
+/**
+ * Walks all <img> elements in {@code container} and binds click → open
+ * lightbox. Safe to call repeatedly: already-bound elements are skipped.
+ *
+ * @param container any HTMLElement that holds rendered markdown
+ */
+function attach(container: HTMLElement | null) {
+  if (!container) return
+  const images = Array.from(container.querySelectorAll('img'))
+  if (images.length === 0) return
+
+  // Snapshot URLs so the order is stable for the duration of this view.
+  const newUrls = images
+    .map(img => img.getAttribute('src'))
+    .filter((u): u is string => !!u && u.length > 0)
+  urls.value = newUrls
+
+  images.forEach((img, idx) => {
+    if (img.dataset[BOUND_FLAG] === 'true') return
+    img.dataset[BOUND_FLAG] = 'true'
+    img.style.cursor = 'zoom-in'
+    img.addEventListener('click', (event) => {
+      event.preventDefault()
+      activeIndex.value = idx
+      open.value = true
+    })
+  })
+}
+
+defineExpose({ attach })
+</script>
diff --git a/mateclaw-ui/src/views/Wiki/components/WikiPageViewer.vue b/mateclaw-ui/src/views/Wiki/components/WikiPageViewer.vue
index 90e488b9..5ce06cfb 100644
--- a/mateclaw-ui/src/views/Wiki/components/WikiPageViewer.vue
+++ b/mateclaw-ui/src/views/Wiki/components/WikiPageViewer.vue
@@ -46,9 +46,17 @@
     </div>
 
     <!-- Content -->
-    <article v-if="!editing" class="page-content markdown-body" v-html="renderedContent"></article>
+    <article
+      v-if="!editing"
+      ref="articleRef"
+      class="page-content markdown-body"
+      v-html="renderedContent"
+    ></article>
     <textarea v-else v-model="editContent" class="page-editor" rows="30"></textarea>
 
+    <!-- Click-to-zoom overlay for inline images. Bound after each render via attach(). -->
+    <ImageLightbox ref="lightboxRef" />
+
     <!-- RFC-033: Related Pages Panel (replaces backlinks) -->
     <RelatedPagesPanel
       v-if="!editing && store.currentKB"
@@ -88,7 +96,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, watch, onMounted } from 'vue'
+import { ref, computed, watch, onMounted, nextTick } from 'vue'
 import { useI18n } from 'vue-i18n'
 import { useWikiStore, isProtectedPage, type WikiPage } from '@/stores/useWikiStore'
 import { wikiApi } from '@/api/index'
@@ -97,6 +105,7 @@ import { Link, SetUp } from '@element-plus/icons-vue'
 import PageHeader from './PageHeader.vue'
 import RelatedPagesPanel from './RelatedPagesPanel.vue'
 import CitationDrawer from './CitationDrawer.vue'
+import ImageLightbox from './ImageLightbox.vue'
 
 const { t } = useI18n()
 const store = useWikiStore()
@@ -108,6 +117,10 @@ const backlinks = ref<WikiPage[]>([])
 const citationDrawerOpen = ref(false)
 const enrichToast = ref('')
 
+// Refs for the lightbox post-render binding step.
+const articleRef = ref<HTMLElement | null>(null)
+const lightboxRef = ref<{ attach: (el: HTMLElement | null) => void } | null>(null)
+
 // RFC-051 PR-8: protection state for delete-button gating + badge rendering.
 const isSystem = computed(() => store.currentPage?.pageType === 'system')
 const isProtected = computed(() => isProtectedPage(store.currentPage))
@@ -131,6 +144,15 @@ const renderedContent = computed(() => {
   return renderMarkdown(content)
 })
 
+// Bind the image lightbox to the rendered article on every content swap.
+// Awaits a microtask so v-html has a chance to repopulate the DOM, then
+// asks the lightbox to walk <img> tags and attach click handlers. Already-
+// bound elements are skipped by the lightbox itself.
+watch(renderedContent, async () => {
+  await nextTick()
+  lightboxRef.value?.attach(articleRef.value)
+})
+
 watch(() => store.currentPage, async (page) => {
   if (page && store.currentKB) {
     editing.value = false