publish-seo-batch

Publish SEO Batch

The daily content pipeline: source topics -> dedup slugs -> write posts -> register -> build -> commit -> deploy -> prove live -> ping IndexNow. Default batch size is 10 unless the user says otherwise. Every step has a check; a batch is not "published" until step 8 passes, then step 9 nudges the search engines.

Step 0: State check

git -C /Users/promode/qaskills status --short   # note pre-existing WIP; you will NOT stage it
date +%F                                        # today's date, used in every post and the commit message

Pre-existing modified/untracked files are the user's WIP. Leave them alone.

Step 1: Source topics

Saved GSC reports in docs/seo/KEYWORD-OPPORTUNITIES-*.md are historical and fully published; do not source topics from them. Instead:

1. If the user supplied topics or keywords, use those.
2. Otherwise WebSearch for net-new opportunities in the site's clusters: Playwright (releases, features, integrations), LLM evaluation (DeepEval, Ragas, promptfoo, Langfuse), API testing, load testing, AI test generation, agentic testing, CI/CD. Favor high-intent long-tail: "X vs Y 2026", "how to X", tool + new-version guides, fresh industry reports.
3. Cross-check each candidate against existing coverage (step 2). Prefer gaps over rewrites.
4. If the research produced a reusable keyword dataset, save it as docs/seo/KEYWORD-OPPORTUNITIES-YYYY-MM.md and include it in the commit.

Step 2: Dedup every slug (MANDATORY before writing any file)

Batch arrays are spread into the registries LAST, so a colliding slug silently replaces a real article. This has shipped a stub over a full article before.

cd /Users/promode/qaskills
for s in slug-one slug-two; do
  hits=$(grep -rn "$s" packages/web/src/app/blog/posts/ | grep -v "$s-something-longer" | wc -l)
  echo "$s: $hits hits"
done

Any hit (filename, posts map key, postList entry, batch array entry): choose a different slug or drop the topic. Also scan titles in posts/index.ts for near-duplicate topics; a second article on the same query cannibalizes the first.