Korean Naver Blogger Discovery

Find Naver bloggers who post deeply and repeatedly about a given KRX stock. Feeds kr-naver-insight with a ranked candidate list. The goal is to surface independent long-form voices — not to scrape personal data.

When To Use

• Before writing a full memo, when the user wants independent viewpoints in the Street / Alternative Views section.
• Before a refresh pass, to check whether new bloggers have started covering a stock the user is tracking.
• As input to kr-naver-insight, which reads the output JSON and fetches the actual post bodies.

Do not use for:

• Sell-side broker list discovery — use DART/news skills instead.
• ETF or sector-wide coverage — bloggers cluster per stock, not per sector.
• Any task where the user does not need Naver voices specifically.

Workflow

1. Read the request. Confirm company + ticker + optional --min-posts.
2. Check cache. If .tmp/naver-blog-cache/bloggers/<ticker>/<YYYY-MM-DD>.json exists for today, prefer it unless --no-cache is set.
3. Build dynamic queries via build-query-set.js:
   • Base: <company> 투자 분석, <company> 실적
   • DART products (if --dart-file): top nouns from 사업의 내용 section
   • News trends: top keywords from Naver News headline frequency analysis
   • Context (if --context-file): terms from 핵심 논점 / 투자 포인트 sections
   • Capped at 8 queries. Use --no-auto-queries for legacy 3-query mode.
4. Pass 1 — Keyword search. Run every query through kr-naver-browse.searchNaverBlog and accumulate hits by blogId.
5. Pass 2 — Nickname anchor. Call fetchRelatedQueries(company) to scrape the blog search page's 연관검색어 block. Strip generic tails (주가, 전망, 배당, ...) via extractNicknameQueries. Fire each surviving suggestion as a searchNaverBlog call; hits are tagged with the surfacing nickname (nicknameHitCount, nicknameAnchors).
6. Pass 3 — Roundup-post body mining. Scan all pass 1 + pass 2 hits for titles matching isRoundupTitle (N인 N색, bracketed nickname lists). Fetch up to 3 matching posts' full page text via browseText and call extractBlogIdRefs to pull every blog.naver.com/<blogId> reference out of the body. Each referenced blogId is injected as an anchored candidate with roundupMentioned: true.
7. Select candidates: keyword-only candidates are capped at --max-candidates; anchored candidates (nickname OR roundup) are uncapped.
8. Per-blog search each candidate via PostSearchList.naver. Count posts whose title contains the company name or ticker (relevantPostCount). Classify each match as trading / listicle / dedicated via the shared title-filter lib. dedicatedPostCount is the primary depth signal.
9. Detect trading-shop blogs: if 50%+ of the blog's page-1 post titles are TA-flavored, mark isTradingBlog: true and exclude — including anchored candidates. A trading shop is a trading shop.
10. Rank by anchored-status desc → dedicatedPostCount desc → inBlogPaginationPages desc.
11. Qualify: anchored OR dedicatedPostCount >= --min-posts, AND isTradingBlog === false. Anchored candidates bypass the min-posts floor because the anchor IS the quality signal.
12. Write a dated JSON file — the schema is documented in references/workflow.md.

Output Contract

• Output path: --output (required). Typical value: .tmp/naver-blog-cache/bloggers/<ticker>/<YYYY-MM-DD>.json.
• JSON shape:

{
  "company": "엘앤에프",
  "ticker": "066970",
  "discoveredAt": "2026-04-12",
  "bloggers": [
    {
      "blogId": "mooyoung_2022",
      "displayName": null,
      "blogTitle": null,
      "subscriberCount": null,
      "relevantPostCount": 6,
      "dedicatedPostCount": 5,
      "tradingTitleCount": 0,
      "listicleTitleCount": 1,
      "isTradingBlog": false,
      "searchHitCount": 1,
      "nicknameHitCount": 4,
      "nicknameAnchors": ["무영"],
      "roundupMentioned": true,
      "roundupSources": ["bosbos2"],
      "inBlogPage1Total": 10,
      "inBlogPaginationPages": 11,
      "latestPostDate": null,
      "categories": [],
      "profileSnippet": null
    }
  ],
  "meta": {
    "searchQueries": 7,
    "candidatesFound": 50,
    "qualified": 8,
    "querySet": { "..." : "..." },
    "relatedQueries": [{ "text": "엘앤에프 무영", "nickname": "무영", "wholeQuery": false }],
    "nicknameQueries": [{ "text": "엘앤에프 무영", "nickname": "무영", "hits": 10 }],
    "nicknameAnchoredCount": 9,
    "roundupFetched": [
      { "sourceBlogId": "bosbos2", "logNo": "224125884833", "refs": ["mooyoung_2022", "naburo", "shmoon305"] }
    ]
  }
}

Fields that cannot be extracted are set to null / [], never fabricated.

Operating Rules

1. Never invent blogger names, subscriber counts, or blog titles. If the page didn't yield it, the field stays null.
2. Respect the 1 s delay between browse requests (handled by kr-naver-browse).
3. Cap keyword-only candidates at --max-candidates (default 10). Anchored candidates (nickname / roundup) are uncapped — they are rare and the whole point of Phase 3 was fixing recall.
4. Never scrape personal info — only public post metadata.
5. Keep the output deterministic given the same inputs: rank by (anchored desc, dedicatedPostCount desc, relevantPostCount desc, inBlogPaginationPages desc, latestPostDate desc, blogId asc).
6. Fail loud if browse-naver.js cannot be required or the browse binary is missing. Do not fall back to a non-Naver source.

Scripts

discover-bloggers.js

node skills/kr-naver-blogger/scripts/discover-bloggers.js \
  --company "엘앤에프" --ticker 066970 \
  --output .tmp/naver-blog-cache/bloggers/066970/2026-04-13.json

Optional flags: --min-posts N, --cache-dir path, --no-cache, --verbose, --queries-file PATH, --dart-file PATH, --context-file PATH, --no-auto-queries, --no-quality-filter.

build-query-set.js

Build a dynamic query set independently (useful for inspection/debugging):

node skills/kr-naver-blogger/scripts/build-query-set.js \
  --company "엘앤에프" --ticker 066970 --output /tmp/queries.json --verbose

Optional flags: --dart-file PATH, --context-file PATH, --max-queries N.