VelimirMueller Audit Content Quality
Score first, then fix only the gaps — never rewrite what already passes. The procedure is generic; the law comes from a rubric file, so a team's own scorecard (points, knockouts, house rules) plugs in without changing the skill.
1. Gate — is this a public page?
Run the gate from ../_shared/page-types.md. Content quality scoring is for pages meant
to be found and read; app surfaces are out of scope.
2. Load the rubric
Per ../_shared/rubric-convention.md:
cat .claude/rubrics/content-quality.md 2>/dev/null # project law, if present
Present → it is the entire rubric (criteria, knockouts, any points/weights it defines).
Present but empty — or with no ## criterion headings — → malformed: report it and
stop; do not fall through to the default (../_shared/rubric-convention.md).
Absent → use ./default-rubric.md, and offer the install at the end of the report.
3. Gather the page
curl -s "$URL" -o /tmp/page.html # or the built file
Read the rendered text top to bottom — the audit quotes evidence, so skimming isn't enough. Identify the page's target question/intent first (from title + h1); several criteria score against it.
4. Score
Walk the rubric criterion by criterion:
- Knockouts (
[K.O.]) first. Any knockout failure fails the page outright — report it, fix it, re-run before scoring the rest in detail. - Each criterion gets a verdict (pass/fail, or the rubric's own scale) and quoted evidence from the page. No verdict without a quote — evidence is what makes the audit repeatable.
5. Report
| Criterion | Verdict | Evidence |
|---|---|---|
| [K.O.] Content in served HTML | PASS | view-source: h1 + body text present |
| Direct answer up front | FAIL | opening paragraph is a 140-word anecdote; answer arrives in §3 |
| ... | ... | ... |
Close with: the failures in fix order (knockouts → cheapest wins → rest), and — if the
bundled default was used — the offer to install it to .claude/rubrics/content-quality.md.
6. Fix only the failures
Apply the smallest edit that turns each failed criterion green: move the answer paragraph up, add the byline/date block, add the two internal links. Touch nothing that passed. Re-run the audit; the page now passes, and a second run changes nothing — idempotence is the proof the audit is honest.
References
- ./default-rubric.md — the bundled scorecard (installable as the project's starting point).
- ../_shared/rubric-convention.md — lookup order, the
[K.O.]format, the install offer. - ../set-up-seo/SKILL.md — the technical baseline (crawlability, metadata, schema) this audit assumes.