| name | content-review |
| description | **WORKFLOW SKILL** — Review new/changed Jekyll content (Markdown under pages/**) for SEO, consistency, polish, accessibility, and technical accuracy, using PER-COLLECTION thresholds. USE FOR: reviewing a content PR before merge, auditing a post/doc/quickstart/note you just wrote, pre-publish SEO + style checks, resolving AI-content-review feedback. INVOKES: scripts/content-review.rb (deterministic tier), the content-reviewer Claude Code agent, .github/config/content_review.yml, the per-collection .github/instructions/*.instructions.md. DO NOT USE FOR: code/theme review (use /code-review), bulk front-matter-only normalization (use /frontmatter-maintainer), or releases (use commit-publish). |
Content Review
Run the Zer0-Mistakes two-tier content review on whatever a branch/PR changes. Tier 1 is deterministic (no API key); tier 2 is the editorial Claude Code agent. Thresholds and authoring rules are resolved per collection — docs follow the documentation guidelines, posts are graded as articles, notes as short-form, etc.
When to use
- Reviewing a PR that adds or edits content under
pages/**.
- Before publishing a post/doc/quickstart/note/quest you just drafted.
- Resolving feedback from the
ai-content-review.yml workflow.
Inputs (source of truth)
| File | Provides |
|---|
.github/config/content_review.yml | Per-collection SEO/quality thresholds, scoring, focus, assigned skills/prompts |
.github/config/frontmatter_schema.yml | Required front-matter fields per collection |
.github/instructions/content-review.instructions.md | Baseline authoring + resolution rules |
per-collection instructions (e.g. documentation.instructions.md) | Collection-specific rules (named in each file's review output) |
Pipeline
1. Deterministic tier (always — no secrets)
ruby scripts/content-review.rb --changed --base origin/main \
--json /tmp/content-review.json --summary /tmp/content-review.md
ruby scripts/content-review.rb --files "pages/_posts/2026-…-foo.md"
Each result carries collection, per-collection score/fail_under, the governing instructions array, and categorized issues. Stop and read the JSON before reasoning further — don't re-derive the mechanical checks.
2. Editorial tier (Claude Code agent)
Delegate the judgment layer to the content-reviewer agent (or the /content-review prompt). For each file, it reads the instructions paths the deterministic tier listed, then reviews: SEO/AIEO, consistency, polish, accessibility, technical accuracy.
3. Report
Use the agent's output format — verdict (✅ approve / 💬 comment / 🔧 request changes), per-file findings tagged 🔴 must-fix · 🟡 should-fix · 🔵 nice-to-have, each with a concrete suggested rewrite. Reviews advise; they don't auto-block (strictness.ci: warn).
Resolving findings (order)
- Front matter (required fields, list-vs-string, ISO dates)
- SEO (title/description length, keywords)
- Structure (headings, code-fence languages, alt text)
- Prose (clarity, consistency, accuracy)
Re-run ruby scripts/content-review.rb --changed after fixing, and always bump lastmod on any file you edit.
Reporting back
| Tier | Status |
|---|
| Deterministic (avg score / failing files) | ✅ / ⚠️ |
| Editorial findings (🔴 / 🟡 / 🔵 counts) | … |
| Verdict | ✅ / 💬 / 🔧 |