ai-seo-articles（AI SEO articles 搜索）

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.

To use this skill for a different project:

1. Create config/[project-name]/ with subdirectories:

- markets/market-a/COPY.md + CLUSTERS.md - markets/market-b/COPY.md + CLUSTERS.md - markets/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)

1. Update the "Active Project" table above to point to config/[project-name]/
2. Create .todo/seo-[project-name]/ with SEO_PROFILE.md, MANIFEST.md, briefs/, drafts/
3. Update ORCHESTRATOR.md Phase 0 config file paths to reference the new config and runtime paths

How to Invoke

Execution

For run seo commands:

1. Read .cursor/skills/ai-seo-articles/prompts/ORCHESTRATOR.md — the full 6-phase pipeline
2. The orchestrator reads config files (Phase 0) before any other action
3. 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)

Directory Index

Runtime Data (in.todo/seo/ — not in this skill)

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.mjs must 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

1. Articles published (count + full production URLs, e.g. https://blink.new/blog/[slug])
2. Clusters worked + new coverage scores
3. One M2 backlink opportunity flagged for human action
4. What to prioritise next run
5. Articles held (with reason and score)