By name: ai-seo-articles description: > AI-powered SEO article production pipeline. Researches keywords, writes high-quality articles with branded hero images and inline visuals, scores quality at 110 pts, processes images via script, and publishes at scale. Configurable for any company via config/ subfolder. Currently configured for: Blink (blink.new). Commands: "run seo", "write seo articles", "seo run", "run seo 8 articles", "run seo b7 sprint", "run seo [market] focus".
ai-seo-articles — Master Skill
One skill, one source of truth for AI-powered SEO article production. Configurable for any project.
Active Project
Change this section to switch projects. Everything else reads from these paths automatically.
| Setting | Value |
|---|---|
| Project | Blink (blink.new) |
| Config path | .cursor/skills/ai-seo-articles/config/blink/ |
| Runtime data | .todo/seo/ (SEO_PROFILE.md, MANIFEST.md, briefs/, drafts/) |
To use this skill for a different project:
- Create
config/[project-name]/with subdirectories:markets/market-a/COPY.md+CLUSTERS.mdmarkets/market-b/COPY.md+CLUSTERS.mdmarkets/market-c/COPY.md+CLUSTERS.md(add more markets as needed)COMPANY.md(identity, CDN, brand voice, global rules)IMAGES.md(character system + scene table)
- Update the "Active Project" table above to point to
config/[project-name]/ - Create
.todo/seo-[project-name]/with SEO_PROFILE.md, MANIFEST.md, briefs/, drafts/ - Update ORCHESTRATOR.md Phase 0 config file paths to reference the new config and runtime paths
How to Invoke
| Command | What runs |
|---|---|
run seo |
Standard run — 5 articles across top 5 priority clusters |
run seo 8 articles |
Extended run — 8 articles |
run seo b7 sprint |
5-6 "How to Build X" articles + 2-3 other priorities |
run seo claude code focus |
Prioritise Market C clusters (C1-C8) |
run seo openclaw focus |
Prioritise Market A clusters (A7, A9, A1) |
run seo vibe coding focus |
Prioritise Market B clusters (B2, B4, B7, B8, B9) |
audit blog |
Scan all published articles for MDX-breaking issues (~30s) |
fix blog |
Auto-fix all broken articles (strip HTML comments, fix images) |
Execution
For run seo commands:
- Read
.cursor/skills/ai-seo-articles/prompts/ORCHESTRATOR.md— the full 6-phase pipeline - The orchestrator reads config files (Phase 0) before any other action
- Execute the orchestrator phases 0-6 exactly as written
For audit blog:
node .cursor/skills/ai-seo-articles/scripts/audit-fix-blog.mjs
Lists all broken articles. Exit 0 = all clean, exit 2 = issues found.
For fix blog:
# Fast — strip HTML comments (recommended default):
node .cursor/skills/ai-seo-articles/scripts/audit-fix-blog.mjs --fix-strip
# Full — generate replacement images + strip comments:
node .cursor/skills/ai-seo-articles/scripts/audit-fix-blog.mjs --fix
Scans all 200+ articles, auto-fixes broken ones, publishes the fixes.
Re-run audit afterwards to confirm Broken: 0.
Config Files (what each contains)
| File | Contains |
|---|---|
config/blink/COMPANY.md |
Company identity, CDN config, global CTA rules, brand language dos/don'ts |
config/blink/IMAGES.md |
CHARACTER ANCHOR, scene table by article type, 6 validated prompts, inline image system |
config/blink/markets/openclaw/COPY.md |
Market A: CTAs, talking points, pricing anchors, competitor positioning |
config/blink/markets/openclaw/CLUSTERS.md |
Market A: A1-A10 cluster definitions and keywords |
config/blink/markets/vibe-coding/COPY.md |
Market B: CTAs, talking points, pricing anchors, competitor positioning |
config/blink/markets/vibe-coding/CLUSTERS.md |
Market B: B1-B11 cluster definitions and keywords |
config/blink/markets/cursor-claude/COPY.md |
Market C: CTAs, happy path, builder section template, style rules |
config/blink/markets/cursor-claude/CLUSTERS.md |
Market C: C1-C8 cluster definitions and keywords |
Directory Index
| Path | What it contains |
|---|---|
SKILL.md |
This file. Entry point with Active Project config section. |
config/blink/COMPANY.md |
Company identity, CDN, global brand + CTA rules |
config/blink/IMAGES.md |
Clay character system, scene table, 6 example prompts |
config/blink/markets/openclaw/COPY.md |
Market A copy: CTAs, talking points, pricing, competitors |
config/blink/markets/openclaw/CLUSTERS.md |
Market A clusters: A1-A10 keywords and strategy |
config/blink/markets/vibe-coding/COPY.md |
Market B copy: CTAs, talking points, pricing, competitors |
config/blink/markets/vibe-coding/CLUSTERS.md |
Market B clusters: B1-B11 keywords and strategy |
config/blink/markets/cursor-claude/COPY.md |
Market C copy: CTAs, happy path, builder template, style rules |
config/blink/markets/cursor-claude/CLUSTERS.md |
Market C clusters: C1-C8 keywords and strategy |
prompts/ORCHESTRATOR.md |
The brain. Generic 6-phase pipeline — loads config at Phase 0 |
reference/VOICE.md |
Writing voice. 10 writing rules with real examples + blueprint cheat sheet. Read second (after brief, before SOP.md). |
reference/SOP.md |
SOP. Production workflow, article type templates, frontmatter schema, MDX component allowlist, SEO/external-link rules, publishing checklist |
reference/GEO.md |
GEO + technical SEO. AI citation optimization, schema markup, robots.txt, sitemap, llms.txt, verification commands |
scripts/validate-draft.mjs |
Run before publishing. Validates a draft: frontmatter, images, MDX safety, word count, stats, links. Exit 0=pass, 2=warnings, 1=fail |
scripts/process-inline-images.mjs |
Run after writing. Replaces INLINE_IMAGE comments with CDN-hosted images. Exit 0=processed, 2=no slots found, 1=fatal |
scripts/audit-fix-blog.mjs |
Blog health guard. Scans all CMS articles for MDX-breaking patterns, bulk auto-fixes. Run after every SEO run. |
scripts/audit-seo-health.sh |
Technical audit. Checks noindex, robots.txt, sitemap, schema, llms.txt, OG images |
scripts/check-competitor-serps.sh |
SERP monitoring. Checks SERP position for key keywords |
Runtime Data (in .todo/seo/ — not in this skill)
| Path | Purpose | Who writes it |
|---|---|---|
.todo/seo/SEO_PROFILE.md |
Markets, clusters, Global Priority Stack, live scores | Human + orchestrator |
.todo/seo/MANIFEST.md |
Run history — append after every run | Orchestrator |
.todo/seo/briefs/BRIEF_[slug].md |
Per-article research briefs | Research worker |
.todo/seo/drafts/DRAFT_[slug].mdx |
Article drafts in-flight | Writer worker |
Quick Scripts
# Validate a draft before publishing (checks frontmatter, images, MDX safety, links):
node .cursor/skills/ai-seo-articles/scripts/validate-draft.mjs .todo/seo/drafts/DRAFT_[slug].mdx
# Process inline images (run after writing, before publishing):
node .cursor/skills/ai-seo-articles/scripts/process-inline-images.mjs .todo/seo/drafts/DRAFT_[slug].mdx
# Technical SEO audit:
bash .cursor/skills/ai-seo-articles/scripts/audit-seo-health.sh
bash .cursor/skills/ai-seo-articles/scripts/audit-seo-health.sh http://localhost:3000 # against localhost
# SERP presence check:
bash .cursor/skills/ai-seo-articles/scripts/check-competitor-serps.sh
# Blog health audit (run after EVERY publish — mandatory Phase 5b gate):
node .cursor/skills/ai-seo-articles/scripts/audit-fix-blog.mjs # audit: list broken articles
node .cursor/skills/ai-seo-articles/scripts/audit-fix-blog.mjs --fix-strip # fast fix: strip HTML comments
node .cursor/skills/ai-seo-articles/scripts/audit-fix-blog.mjs --fix # full fix: generate images + strip
Hard Stops (always enforced)
- B11 Customer Stories: never write without real customer data provided by a human
- B10 AI Comparisons: cap at 5 total; never prioritise over P1-P14
- Score gate: nothing publishes below 90/110; three revision passes max, then hold
- No duplicate keywords: check MANIFEST.md before any brief
- No fabricated data: invented stats or quotes → editor rejects, writer rewrites with sources
- Slug rules: no year suffix (
-2026). Timeless slugs. Exception: changelogs only - No manual CTAs: auto-injected by rendering engine. DO NOT write CTA sections.
- Zero HTML comments in published MDX:
audit-fix-blog.mjsmust exit 0 (Broken: 0) before any run closes. HTML comments crash every page visitor. Note:<!-- INLINE_IMAGE_* -->slots are draft-only placeholders — the publish pipeline converts them to images before CMS. They must never appear in published content.
End-of-Run Report Format
- Articles published (count + full production URLs, e.g. https://blink.new/blog/[slug])
- Clusters worked + new coverage scores
- One M2 backlink opportunity flagged for human action
- What to prioritise next run
- Articles held (with reason and score)