diff --git a/.claude/skills/mt-add-image/SKILL.md b/.claude/skills/mt-add-image/SKILL.md
index f4feef2..83a8834 100644
--- a/.claude/skills/mt-add-image/SKILL.md
+++ b/.claude/skills/mt-add-image/SKILL.md
@@ -28,13 +28,21 @@ Find the source post:
 ```bash
 node .claude/skills/mt-add-image/scripts/find-substack-post.js --uuid <uuid>
 ```
-- `found: true` → label = `caption` if non-empty, else `postTitle`.
-- `found: false` → retry with the deeper sitemap crawl (slower — scans ~3 months, warn the user it may take ~15s):
+- `found: false` → retry with the deeper sitemap crawl (slower — scans ~3 months; warn the user it may take ~15s):
   ```bash
   node .claude/skills/mt-add-image/scripts/find-substack-post.js --uuid <uuid> --deep
   ```
-  - `found: true` → label as above.
-  - `found: false` → no automatic label; go to step 3 (ask). The result reports `scanned` + `cutoff` — mention how far back it looked.
+  The result reports `scanned` + `cutoff` — mention how far back it looked.
+- `found: false` after `--deep` → no source post; go to step 3 (ask) and/or step 4 (add publication).
+
+**When `found: true` — pick the label (confirm-from-candidates):**
+ByteByteGo doesn't attach captions to images (verified: empty `alt`, no `<figcaption>`), and image order can't be mapped to titles automatically because sponsor/video items interleave. So the script returns `candidates` — the issue's topic titles from the post's TOC. Pick the label this way:
+1. If `caption` is non-empty (some publications do caption images) → propose it as the default.
+2. Otherwise present `candidates` to the user via `AskUserQuestion` (the correct title is in this list):
+   - "Image is from **[postTitle]** (`postUrl`). Which title matches it?"
+   - Offer the candidates. If there are more than 4, show them as a numbered list in text and ask the user to pick a number (AskUserQuestion allows max 4 options).
+   - Always include a "Type my own" escape.
+3. Label = the user's pick (or the typed value). Localize to Vietnamese where natural; keep proper nouns.
 
 ### 2b. Non-Substack image (`isSubstack: false`)
 Best-effort only (no reverse-image-search): if you happen to know the page that contains the image, WebFetch it and read the OpenGraph title / nearby caption. Usually there's no containing page from just an image URL → go to step 3.
@@ -58,13 +66,13 @@ Create `### Bonus` / `**Images:**` if missing.
 ```
 ✅ Image added to Newsletter #[number] (Bonus → Images)
 📄 content/post/YYYY/MM/DD/index.md
-🏷️  label: "[label]"  (source: rss | sitemap | opengraph | user-input)
+🏷️  label: "[label]"  (source: candidate-pick | caption | user-input; post: [postTitle])
 🖼️  [clean_image_url]
 ```
 
 ## Checklist
 
 - [ ] Source detected (substack + uuid, or non-substack)
-- [ ] Label resolved by priority (caption → post title → your input)
+- [ ] Label confirmed (caption if present, else user picked from candidates, else typed)
 - [ ] Entry under Bonus → **Images** (section/subsection created if missing)
 - [ ] Label localized to Vietnamese where natural (keep proper nouns)
diff --git a/.claude/skills/mt-add-image/scripts/find-substack-post.js b/.claude/skills/mt-add-image/scripts/find-substack-post.js
index 88f0d78..c4380fb 100644
--- a/.claude/skills/mt-add-image/scripts/find-substack-post.js
+++ b/.claude/skills/mt-add-image/scripts/find-substack-post.js
@@ -76,6 +76,30 @@ function itemLink(item) {
   return m ? m[1].trim() : "";
 }
 
+// Extract candidate topic titles from a post's TOC bullet list. ByteByteGo does
+// not attach captions to images — the topic titles live only in the "in this
+// issue" bullets. We can't reliably map image→title automatically (sponsor /
+// video items interleave and break positional order), so we surface these
+// candidates for the user to pick from. Light filtering keeps the list short:
+// dedupe, drop sub-point explanations ("Term: long sentence") and over-long
+// lines. The correct title is always present; the user selects it.
+function extractCandidates(html) {
+  const seen = new Set();
+  const out = [];
+  const re = /<li[^>]*>\s*(?:<p[^>]*>)?([\s\S]*?)(?:<\/p>)?\s*<\/li>/g;
+  let m;
+  while ((m = re.exec(html))) {
+    const text = stripTags(m[1]);
+    if (!text) continue;
+    if (text.length < 6 || text.length > 70) continue;     // titles are short; long lines are sub-point explanations
+    const key = text.toLowerCase();
+    if (seen.has(key)) continue;                           // content is duplicated in the page
+    seen.add(key);
+    out.push(text);
+  }
+  return out;
+}
+
 // If the UUID sits inside a <figure>…</figure>, return that figure's
 // <figcaption> text. Cover images live in <enclosure> (no figure) → "".
 function captionForUuid(item, id) {
@@ -134,6 +158,7 @@ async function searchSitemap(publication, id, maxFetch = 40, monthsBack = 3) {
         postTitle: postTitleFromHtml(html),
         postUrl: c.url,
         caption: captionForUuid(html, id),
+        candidates: extractCandidates(html),
       },
       scanned,
       cutoff: cutoff.toISOString().slice(0, 10),
@@ -157,6 +182,7 @@ async function searchRss(publication, id) {
       postTitle,
       postUrl: itemLink(item),
       caption,
+      candidates: extractCandidates(item),
     };
   }
   return null;