WORKPULSION SYNTHETIC EDITORIAL TRIALS
October 12, 2026

The two procedures were manually applied in a Codex session. No plugin installation or Claude Code runtime test. No baseline comparison or conversion measurement.

Marketing input: marketing/.agents/product-marketing.md
Marketing output: marketing/output.txt
Website input and correction: review/before.html and review/after.html
Our checker: review/check.py (not a third-party script)
Recorded results: results.json

To repeat the two website checks with Python, from the review folder run:
python check.py before.html
Expected: exit 1, wrong canonical and missing Contact target.
python check.py after.html
Expected: exit 0, both limited checks pass.

From the bundle root, python run.py repeats both checks and checks four aspects of the supplied marketing draft. It does not generate new copy or invoke any skill.

Exact source revisions reviewed:
{
  "coreyhaines31/marketingskills": "1efedbc5148b54b2f0f6c6c9fe0be62e151c7fff",
  "obra/superpowers": "bb92a77741419a4ab5f06e711a283343f1ada0c3",
  "anthropics/skills": "dbd4588f9e1033efb41dad4bef2f7947c8993d44",
  "AgriciDaniel/claude-seo": "4b99de2f7de7e7d5247042e5fb5b4ea9368ef734"
}

Third-party sources are linked in the article, not redistributed in this bundle.
