SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsgr-seo-patrol

gr-seo-patrol

Marketing & growth

View repository

Updated 10-15 21:52

About this skill

gr-seo-patrol — SEO/GEO daily patrol

When to use

Scenario Action
"Run today's SEO daily report" Run the full patrol (scripts/daily-report.py)
"A keyword's ranking dropped" Single-word diagnosis (Claude directly executes the diagnosis flow below)
"Fix canonical" Batch merge (scripts/canonical-fix.py)
"Rescue this article" Social media rescue (scripts/rescue-post.py) — title front-loading + 3 high-authority internal links
"Check GA4 deployment" Scan HTML for G-XXXXX tag
"Check llms.txt" HTTP 200 + byte count

Core flow

1. Daily report mode

Input: none (fully automatic) or a specified keyword list Output:

  • Keyword ranking diff table (today vs yesterday's baseline)
  • Google index count
  • llms.txt status
  • GA4 coverage status
  • 🚨 Anomaly alerts (dropped out of top 100, overall refresh)

2. Single-word diagnosis

Input: 1 keyword Process:

  1. Run SERP for 4-5 long-tail variants
  2. Check whether the target URL returns HTTP 200
  3. Use site: query to confirm whether it is indexed
  4. Compare with data from 3 days ago (if history exists)
  5. 2026-06-24 newly added checks：
    • Confirm the keyword is in the volume 300–1,000 / KD 5–35 range (if not, mark as low priority)
    • Check whether the page title contains best/free/top/guide+year (CTR formula)
    • Check whether the page's internal link density exceeds 2-3/paragraph
    • Confirm the page has BreadcrumbList schema
    • Confirm the last-updated date is displayed (freshness signal) Output: choose one of avalanche / individual / SERP refresh / dead link, with a 2026 compliance score

3. Canonical batch fix

Input: N articles on the same topic + 1 master URL Process: replace the canonical_url: → master in each article's frontmatter, GitHub API PUT Safety guardrails:

  • Only modify _posts/ .md files under
  • Each article is committed separately (for easy rollback)
  • Skip ja/ko (hreflang replacement, keep self-canonical)

4. Social media rescue

Input: 1 avalanche URL Actions:

  • a. Rewrite title: front-load the primary keyword, length ≤ 60 characters,must contain best/free/top/guide+year(2026-06-24 CTR formula)
  • b. Rewrite meta description:keyword rephrase(e.g. "figma alternative" → "the alternative to Figma")
  • c. Inject internal links from top-3 high-authority articles (with anchor text = target keyword),max 2-3 internal links per paragraph
  • d. Proactively add 1-2 competitor blog external links(raise trust score → raise ranking)
  • e. Check and add/update the [Last Updated Date] (freshness signal)
  • f. Confirm the BreadcrumbList schema is configured
  • g. Log the changes before and after

API dependencies

Service Env var Purpose
DataForSEO DATAFORSEO_B64 SERP query
GitHub PAT GITHUB_TOKEN read/write _posts/

For the full key template, see docs/api-keys-template.md。

SERP query template

import urllib.request, json
def serp(kw, loc=2840, lang="en", depth=100):
    key = os.environ["DATAFORSEO_B64"]
    payload = json.dumps([{"keyword": kw, "location_code": loc,
                           "language_code": lang, "device": "desktop",
                           "depth": depth}]).encode()
    req = urllib.request.Request(
        "https://api.dataforseo.com/v3/serp/google/organic/live/advanced",
        data=payload,
        headers={"Authorization": f"Basic {key}",
                 "Content-Type": "application/json"})
    with urllib.request.urlopen(req, timeout=40) as r:
        return json.loads(r.read())

Common location_code:

  • 2840 US / 2826 UK / 2392 JP / 2410 KR / 2156 CN (invalid, China does not return)

Execution scripts

The actually executable scripts are under scripts/ :

  • daily-report.py — full daily report
  • canonical-fix.py — canonical bulk merge
  • rescue-post.py — social media article rescue
  • diag-keyword.py — single-keyword diagnosis (roadmap, temporarily replaced by the "single-word diagnosis" process below)

Each script can be run independently. When invoking, prefer the Bash tool; do not rewrite the scripts.

Output specifications

  1. Give the table first, then the conclusion
  2. Use arrows for differences：#6 ↑ / #19 ↓↓
  3. More than 3 anomalies → open a separate "🚨 Anomalies" section
  4. All rankings must include location_code, do not assume

Cascade recommendations

  • Found cannibalization (multiple articles ranking close together for the same keyword) → gr-blog-post do canonical consolidation
  • Found N keywords dropped out of the top 100 on the same day → wait 24h first, do not force content changes
  • Found new keyword opportunities (long-tail top 30) → gr-blog-post expand

Anti-patterns

  • ❌ Do not use grep / find to scan _posts/, use the GitHub Contents API instead
  • ❌ Do not directly git push in the production environment —— use the Contents API PUT
  • ❌ Do not serp 50 keywords at once —— too large a batch easily triggers rate limit, do batches of 6-10
  • ❌ After changing canonical to master, do not immediately force Google to reindex —— wait 3-7 days for natural crawling

Monthly Full-Site Audit Workflow

Validated 2026-05-07 on 58 pages: caught 43 SERP-truncating titles + 36 schema warns + 27 stop-word slugs in a single 30-min pass. Single layout-level fix (commit 24a0410e) resolved 20 of 43 title issues.

Run once per month before phase2-monthly-checkpoint. Output: HTML report + machine-readable findings.json archived to ~/Downloads/seo-audit/audit-{YYYY-MM-DD}.json.

Stage 1 — Discovery (5 min)

import urllib.request, re
sm = urllib.request.urlopen("https://gingiris.tools/sitemap.xml").read().decode()
urls = [u for u in re.findall(r"<loc>([^<]+)</loc>", sm) if "/blog/" in u]
# typically 50-70 URLs

Stage 2 — Parallel Audit (20 min for 60 pages, 4 threads)

Use the adopted seo-audit-skill scripts in scripts/:

# For each URL — run 2 scripts in parallel
python3 scripts/check-page.py URL --timeout 20    # title, H1, meta, canonical, slug, alt, keyword placement
python3 scripts/check-schema.py URL --timeout 20  # JSON-LD validation

Or batch with Python's concurrent.futures.ThreadPoolExecutor(max_workers=4). Don't go higher than 4 — GitHub Pages CDN throttles aggressive parallel hits.

Each script outputs structured JSON envelope:

{"field": {"status": "pass|warn|fail|info", "detail": "...", "llm_review_required": false}}

Stage 3 — Aggregate Findings

Bucket by category:

  • Title length > 70 chars (SERP truncation risk)
  • H1 length > 70 chars (mobile readability)
  • Meta description < 80 or > 170 chars
  • Schema warns by @type (BlogPosting, Article, Organization)
  • Canonical issues (mismatch with final URL)
  • Slug issues (stop words, uppercase, missing keyword)
  • Image alt text missing on content images

Save aggregated counts + per-issue URL lists to findings.json.

Stage 4 — Layered Fix Strategy (HIGH ROI ORDER)

1️⃣ Layout-level fixes first (1 commit, fixes 20+ pages):

  • Schema bugs in _layouts/default.html
  • Site-name suffix in <title> tag
  • Missing dateModified from last_modified_at frontmatter
  • Organization / Publisher / contactPoint completeness

2️⃣ Config-level fixes (1 commit, fixes site-wide):

  • _config.yml — logo URL (must be absolute), twitter, social, author structure

3️⃣ Per-article batch fixes (1 commit per file, parallelizable):

  • Trim long titles while preserving keyword (target ≤ 70 chars, ideal 50-60)
  • Trim long H1s (target ≤ 70 chars)
  • Expand short meta descriptions (target 120-160 chars)
  • Add missing Citable Statistics blocks for top GSC-impression pages

4️⃣ Skip these (low ROI):

  • Slug stop words (changing breaks 301)
  • Old articles with <50 imp/month (low traffic = low fix priority)

Stage 5 — Verify (after Jekyll rebuild ~60-90s)

Re-run check-schema.py on a sample page. Confirm status: pass for at least: Article · BlogPosting · Organization · FAQPage

Stage 6 — Archive

Commit findings.json to ~/Downloads/seo-audit/audit-{YYYY-MM-DD}.json for trend tracking. Add 2-5 atoms documenting any new lessons learned.

HARD RULE (anti-hallucination guardrail)

Adopted from JeffLi1993/seo-audit-skill — strict whitelist pattern.

⛔ Output ONLY the checks defined in the audit script's JSON envelope.

  • Do NOT add "bonus" checks not in the script output
  • Do NOT contradict the script's status field unless you have additional observable evidence
  • Do NOT invent metrics like "EEAT score 89" — third-party scoring tools are unofficial (per Google's 2026 guidance)
  • Do NOT include checks marked llm_review_required: false in your judgment commentary — the script's status is final
  • If llm_review_required: true, make explicit judgment, document reasoning, then update status

The script envelope is the single source of truth. Treat it as a strict whitelist.

Companion skill

For single-page audits (not full-site), the same scripts power JeffLi1993/seo-audit-skill which produces a polished HTML audit report. Install as a complementary skill if you want client-presentable per-page audits.

Platform Description

EN: Daily SEO/GEO patrol skill (2026-06-24 standards). Tracks SERP keyword rankings (DataForSEO, filters volume 300–1,000 / KD 5–35), checks Google index count, llms.txt accessibility, GA4 tag deployment + AI traffic channel grouping. Rescues underperforming posts by applying title formula (best/free/top/guide+year), keyword rephrase in meta description, internal link density enforcement (≤2–3 per paragraph), competitor blog external link injection, BreadcrumbList schema check, and freshness date update. Fixes canonical issues.

ZH: Daily SEO/GEO patrol skill (2026-06-24 standard). Track SERP rankings (DataForSEO, volume 300–1,000 / KD 5–35 filter), check Google index count, llms.txt reachability, GA4 deployment + AI traffic channel grouping. Rescue avalanche articles: title contains best/free/top/guide+year, meta description keyword rephrase, internal link density ≤2–3/paragraph, proactively cite competitor external links, BreadcrumbList schema, freshness date update, canonical fix.

JA: Daily SEO/GEO patrol skill (2026-06-24 standard). SERP ranking tracking (DataForSEO, volume 300–1,000/KD 5–35 filter), Google index count, llms.txt confirmation, GA4 tag + AI traffic channel setup. Rescue of snow avalanche articles: best/free/top/guide + year in title, meta description keyword rephrase, internal link density ≤2–3/paragraph, competitor blog external link insertion, BreadcrumbList schema, freshness date update, canonical fix.

KO: Daily SEO/GEO patrol skill (as of 2026-06-24). SERP ranking tracking (DataForSEO, volume 300–1,000/KD 5–35 filter), Google index count, llms.txt accessibility, GA4 tag + AI traffic channel group check. Recovery of underperforming posts: apply best/free/top/guide+year to titles, rephrase meta description keywords, internal link density ≤2–3/paragraph, insert external links to competitor blogs, BreadcrumbList schema, update freshness date, fix canonical.