Write Articles That Can Rank

This workflow runs the seo-publish-pipeline skill end to end. It works from two files in your project: content/keyword-backlog.md, a keyword queue built only by vdr research runs, and product.md, the list of features that exist and features that do not. If either file is missing, the agent stops and helps you create it first. Then it picks the oldest pending keyword, classifies the search intent, drafts to a fixed structure, validates every product claim against product.md, runs two anti-slop passes, and scores the draft before you see it. After you publish, it pushes the URL to search engines through IndexNow and marks the backlog entry published.

Paste this into Claude, Codex, Cursor, ChatGPT, or another coding agent. The website, the expected output, and the approval boundary are already written in. First run? Complete the one-time setup; your agent can handle that too.

This is the sequence the agent works through. Read-only steps run on their own; anything that sends or changes something stops and waits for you.

1. 01

   Check the files

   Verify content/keyword-backlog.md and product.md exist; if either is missing, the agent stops and helps you create it.

2. 02

   Build the backlog

   Fill the backlog from research runs only: list_tracked_keywords, keyword_suggestions, and keyword_difficulty verdicts.

3. 03

   Pick and verify

   Take the oldest pending backlog keyword, classify its search intent, and re-verify stale entries against the DR the top 10 demands.

4. 04

   Draft and de-slop

   Draft to the fixed structure, validate every product claim against product.md, run two anti-slop passes, and score before your review.

5. 05

   Submit URL

   After publish, push the URL via IndexNow with the CLI (vdr sites:submit-urls) and mark the backlog entry published.

6. 06

   Check results

   Run get_gsc_performance a few weeks later to check clicks, impressions, and position.

The exact MCP tool calls behind the prompt. Your agent runs these against the verifieddr server; the ones that send anything still stop for your approval.

# The pipeline reads two files (it stops if either is missing):#   content/keyword-backlog.md  <- keyword queue, research runs only#   product.md                   <- features that exist and do NOT # 1. Build the backlog: tracked snapshots first (free)list_tracked_keywords { site: "yourwebsite.com" }keyword_suggestions { domain: "yourwebsite.com" }keyword_difficulty { keyword: "your keyword", domain: "yourwebsite.com" } # 2. Write: oldest pending keyword -> intent -> draft -> score # 3. After publish, push the URL via IndexNow# CLI only: vdr sites:submit-urls https://yourwebsite.com/blog/your-article # 4. Weeks later, check whether the keyword is movingget_gsc_performance { site: "yourwebsite.com", range: "28d" }