Content Reviewer

Content Reviewer

This skill evaluates marketing content drafts against structured quality rubrics. It doesn't line-edit or fix grammar. It assesses whether a piece works at the structural, narrative, and strategic level — and gives actionable feedback to make it stronger.

When to use this skill

• Someone asks you to review, evaluate, or critique a draft
• Someone pastes content and asks "how does this look?" or "is this ready to publish?"
• Someone wants feedback on a blog post, white paper, article, or tutorial
• Someone asks "what would make this better?"
• After using the brand-voice-writer skill to create content, as a quality check

Three rubrics, three content types

Tiger Data's blog content falls into three modes. Each has its own evaluation rubric because they're trying to do fundamentally different things:

Systems Mode (white papers, architectural posts): The goal is to build credibility and shape how people think about a category. The rubric evaluates thesis strength, narrative momentum, category framing, technical authority, selectivity, whether the conclusion feels earned, and memorability. Think: "Would a senior engineer share this?"

Builder Mode (educational posts, tutorials): The goal is to help a developer learn something and take action. The rubric evaluates outcome clarity, practical utility, step-by-step flow, concrete examples, theory discipline, CTA alignment, and builder confidence. Think: "Could someone actually build something after reading this?"

SEO Mode (SEO articles, comparison posts, roundups): The goal is to rank for a target query and serve the searcher's intent while maintaining editorial credibility. The rubric evaluates search intent match, keyword placement, featured snippet optimization, SERP differentiation, internal linking, editorial neutrality (for comparison pieces), and technical SEO readiness. Think: "Would this earn a top-3 ranking and keep the reader from hitting the back button?"

Step 0: Pre-flight check

Read REFERENCES.md from the plugin root and run the pre-flight check described there. Call list_marketing_references() to verify Tiger Den is reachable. If it fails or the tool is not found, STOP — do not continue. Follow the error handling in REFERENCES.md.

Instructions

1. Get the content

The user might paste the draft directly, share a file, or point to a URL. However they provide it, read the full piece before doing anything else.

2. Classify the content type

Pre-classified content: If the prompt that invoked this skill includes Content type: SEO Mode (skip classification), accept that classification and skip directly to Step 3. This happens when seo-article-writer auto-triggers content-reviewer after writing an article — the content type is already known.

Determine which mode this content belongs to. Look at the intent of the piece:

• Systems Mode signals: architectural arguments, "why we built X this way," category-level framing, design philosophy, market landscape analysis, tradeoff discussions
• Builder Mode signals: step-by-step instructions, code examples, "how to do X," tutorials, practical walkthroughs, learning outcomes stated upfront
• SEO Mode signals: the piece targets a specific search query; "X vs Y" or "best X for Y" title format; roundup or listicle structure comparing multiple tools or approaches; content organized around a question someone would type into Google; SERP-oriented structure (FAQ sections, comparison tables, "what is X" definitions); keyword-targeting language in headings

If it's ambiguous between Builder and Systems, ask the user. Some pieces blend both — in that case, note which mode the piece leans toward and evaluate against that rubric, but flag sections where it drifts into the other mode (this drift is usually a problem worth calling out).

If it's ambiguous between SEO and another mode, lean toward SEO if the piece has a clear target keyword and a comparison or roundup structure. SEO articles can contain tutorial elements or architectural arguments, but if the primary purpose is to rank for a search query, classify it as SEO Mode.

If the content is neither mode (e.g., a landing page, email, social post, or one-pager), skip the rubric evaluation and instead review it against the brand voice guide — it has specific tone and structural guidance for each of those content types.

3. Load reference docs and the right rubric

Before evaluating, fetch the reference docs declared in this skill's frontmatter from Tiger Den. Use get_marketing_context to batch-fetch all docs in one call (see REFERENCES.md in the plugin root for details and error handling). If Tiger Den is not connected, do not proceed — tell the user to run /setup.

Fetch all three docs upfront:

• product-marketing-context — for terminology and positioning checks
• educational-content-rubric — the Builder Mode rubric
• white-paper-rubric — the Systems Mode rubric

get_marketing_context(slugs: ["product-marketing-context", "educational-content-rubric", "white-paper-rubric"])

Then use the rubric that matches the content type you classified:

• For Systems Mode: use white-paper-rubric
• For Builder Mode: use educational-content-rubric
• For SEO Mode: read references/seo-article-rubric.md from this skill's directory (local file — no MCP call needed)

4. Check terminology and positioning

Use product-marketing-context to check terminology and positioning accuracy. You don't need this for every review, but do load it if the piece mentions Tiger Data products, makes competitive claims, or positions the product. Wrong terminology or off-message positioning is worth flagging even in a structural review.

5. Run the evaluation

Work through each of the seven dimensions in the rubric. For each dimension, produce the specific outputs the rubric asks for. Be direct and specific — vague feedback like "could be tighter" isn't useful. Point to specific sections, paragraphs, or transitions.

The rubric is designed to surface structural issues, not nitpick. If a dimension is strong, say so briefly and move on. Spend your time on the dimensions where the piece falls short.

Builder Mode dimensions: Outcome Clarity, Practical Utility, Step-by-Step Flow, Concrete Examples, Theory Discipline, Single Clear Next Step, Builder Confidence. For each, the rubric specifies what to output (e.g., for Builder Confidence: where authority is strong, where it feels generic).

Systems Mode dimensions: Core Thesis Strength, Narrative Spine, Category Framing Power, Technical Authority and Credibility, Strategic Selectivity, Conversion Without Selling, Memorability. For each, the rubric specifies what to output (e.g., for Memorability: the core mental model created, whether it's strong enough to shape future thinking).

SEO Mode dimensions: Search Intent Match, Keyword Placement and Coverage, Featured Snippet Optimization, SERP Differentiation, Internal Linking and Content Architecture, Editorial Neutrality (comparison and roundup pieces only — skip for single-topic SEO articles), Technical SEO Readiness. For each, the rubric specifies what to output (e.g., for Editorial Neutrality: whether evidence is symmetric across compared options, where the piece feels promotional rather than editorial).

6. Produce the final assessment

After all seven dimensions, provide:

For Builder Mode:

• A 1–10 rating for Builder Mode quality
• The three highest-impact changes to improve clarity and actionability
• Whether this is truly a tutorial or drifting into thought leadership

Assume the audience is working developers. Avoid hype language. Think in terms of buildability.

For Systems Mode:

• A 1–10 rating for structural quality
• The three highest-impact changes that would elevate it one tier
• A revised high-level outline that would strengthen thesis and momentum

Assume the audience is senior engineers and database architects. Avoid hype language. Think in systems.

For SEO Mode:

• A 1–10 rating for SEO quality
• The three highest-impact changes to improve ranking potential and editorial credibility
• Whether the piece would earn a featured snippet for its target query
• For comparison pieces: whether editorial neutrality is sufficient to build reader trust, or whether the piece reads as a product pitch

Assume the audience is developers evaluating tools. Think in terms of search intent fulfillment and editorial credibility.

For all three modes, the three highest-impact changes should be specific and actionable, not generic advice. The user is going to revise based on your feedback, so prioritize the changes that would move the needle most.

7. Editorial quality check

After the structural rubric, do a quick pass on these editorial fundamentals. These are pass/fail checks — flag issues directly, no scoring needed.

For SEO Mode content: Check 2 (Evidence and specificity) is handled by the SEO rubric's Editorial Neutrality and SERP Differentiation dimensions. Run checks 1, 3, 4, and 5 only.

1. Definition precision — When the piece introduces a key concept, does it define it by its primary function and organizing principle? Flag vague or circular definitions. Sections should open with strong, specific statements that immediately address the topic. 2. Evidence and specificity (Builder and Systems modes only) — Are broad claims backed by concrete data (compression ratios, query speed improvements, cost reductions)? Are real technologies and industry examples named explicitly rather than alluded to generically? 3. Strategic linking — Does the piece link key concepts and product claims to internal resources (docs, case studies, feature pages)? Does it link general technical terms to authoritative external sources? Flag missed linking opportunities. 4. Readability and flow — Are there abrupt transitions or disconnected ideas? After introducing a technical concept, does the piece immediately explain its practical benefit? Flag sections where the reader has to infer the "so what." 5. Terminology consistency — Are architectural labels and product categories used consistently throughout? Flag cases where the same concept gets different names without explanation. (The next step checks whether terminology is correct; this check covers whether it's consistent.) For SEO Mode content: also cross-reference Tiger Data features mentioned in the article against Tiger Docs (if available). Call search_docs(source: "tiger", search_type: "keyword", query: "{feature name}") for each Tiger Data feature. Flag stale product names, outdated API terminology, or incorrect capability descriptions. If Tiger Docs is not reachable, note that terminology was checked against product-marketing-context only.

Keep this section brief. If all five checks pass cleanly, say so and move on.

8. Brand voice spot-check

Separately from the rubric, flag any issues with:

• Terminology: wrong product names, outdated branding (e.g., "Timescale Cloud" instead of "Tiger Cloud"), incorrect capitalization. Check against the glossary in product-marketing-context.
• Em dash verification (automated): Search the content for — (em dash character, U+2014). This is a pass/fail check — any occurrence is a failure. Report the count and locations. This supplements the manual spot-check with a programmatic guarantee.
• Voice violations: em dashes (these are banned), generic AI language, passive voice, hedging ("we believe"), marketing fluff
• Positioning drift: claims that contradict the positioning section in product-marketing-context, feature-first framing instead of problem-first, competitive framing that breaks the guardrails

Keep this section short. If there are no issues, say so. This is a spot-check, not a full brand audit.

9. Offer next steps

After the review, ask whether the user wants:

• A deeper dive on any specific dimension
• Help rewriting specific sections (hand off to brand-voice-writer skill for this)
• A re-review after they've made changes

Tiger Den content enrichment

These features use Tiger Den tools that are already connected (since you fetched reference docs in Step 3).

Content search

Use search_content to find previously published content on the same topic. This gives you useful context: has this topic been covered before? Is this piece retreading old ground or adding something new? Are there existing pieces it should reference or link to?

Voice-match review

If the user mentions who wrote a piece (e.g., "review Matty's draft," "this is Jacky's post," "does this sound like Mike?"):

1. Call get_voice_profile with the author's name to load their writing samples and voice notes 2. Add a Voice Match dimension to the review (in addition to the seven rubric dimensions):

• Compare the draft's sentence rhythm, tone, humor, and paragraph style against the author's profile
• Flag sections that deviate from their natural voice — this often signals over-editing, AI slop, or a ghost-writer who hasn't internalized the author's style
• Note where the voice is strongest (usually the most authentic, least polished sections)

If the user doesn't mention an author, don't load a voice profile — just run the standard rubric.

Calibration notes

A few things to keep in mind when scoring:

• A 7 is good. Most published content from good teams lands in the 6-8 range. A 9-10 means it's genuinely best-in-class — the kind of piece that gets shared widely and referenced months later. Don't grade-inflate.
• The gold standards are 9s. The reference articles linked in the rubrics represent what a 9 looks like. Use them as mental anchors.
• Focus on the highest-leverage feedback. Three strong suggestions beat ten scattered ones. The user is going to revise based on your feedback, so prioritize the changes that would move the needle most.
• Be honest, not harsh. If the piece isn't ready, say so clearly but constructively. "This has a strong core insight but the structure isn't letting it land yet" is better than "This needs major work."

SEO Article Rubric

Evaluation rubric for SEO-driven content: search-intent articles, comparison posts, roundups, and "best X for Y" pieces. Score each applicable dimension 1–10. For each, output specific observations — not just a rating.

A 7 is solid. A 9–10 means it's genuinely strong ranking content that serves the searcher completely and offers something the current top-10 results don't. Don't grade-inflate.

---

Dimension 1: Search Intent Match

What to evaluate: Does the piece correctly identify and serve the searcher's intent? Informational queries need explanatory structure. Commercial investigation queries (comparisons, alternatives, "best X") need evaluation structure — criteria, tradeoffs, a recommendation. Transactional queries need CTAs near the top.

What to output: State the inferred intent type. Assess whether the content structure matches what Google rewards for that intent. Flag mismatches — e.g., a comparison piece that reads like a thought leadership essay, or an informational piece that buries the answer.

Comparison piece note: The piece should serve "X vs Y" or "best X for use case" intent without becoming a product pitch. If the structure funnels the reader toward one option from the opening paragraph, flag it — this reads as marketing, not editorial, and increases bounce rate from searchers expecting a genuine comparison.

---

Dimension 2: Keyword Placement and Coverage

What to evaluate: Is the primary keyword (or its close variants) present in the H1, within the first 100 words, and in at least one H2? Are semantically related terms (LSI keywords — related concepts, synonyms, co-occurring terms) distributed naturally throughout? Is keyword density reasonable — present but not stuffed?

What to output: Confirm or flag keyword placement in H1/intro/H2. Note any obvious keyword gaps or overuse. If the primary keyword is absent from the intro or H1, call it out directly.

Comparison piece note: Competitor and alternative product names should appear naturally and consistently. If a comparison piece avoids naming alternatives explicitly (e.g., always referring to them generically), it loses relevance signals for queries that include those names.

---

Dimension 3: Featured Snippet Optimization

What to evaluate: Does the piece contain structures Google pulls for featured snippets? Definition paragraphs (40–60 words, starting with "X is..."), comparison tables, numbered step lists, and direct-answer paragraphs are the most common snippet types. Are H2s and H3s phrased as questions when the target query is a question?

What to output: Identify the most likely snippet opportunity in this piece (definition, table, list, or step). Assess whether the content is structured to win it. Flag if the best candidate paragraph is too long, buried, or structured in a way that makes extraction difficult.

---

Dimension 4: SERP Differentiation

What to evaluate: Does this piece offer something the current top-10 results for the target query don't? Unique angle, proprietary benchmarks, novel evaluation criteria, fresher data, or a structural format that's more useful. A piece that covers the same ground as existing results in the same order won't displace them.

What to output: Identify what the differentiation claim is (or should be). If the piece doesn't have one, say so — this is often the highest-leverage fix for SEO content that's technically solid but not ranking.

Comparison piece note: Does the piece include evaluation dimensions or use-case distinctions that competitors' comparison pages omit? Unique criteria (e.g., evaluating tools on operational complexity, not just performance) are strong differentiators.

---

Dimension 5: Internal Linking and Content Architecture

What to evaluate: Does the piece link to relevant product pages, documentation, tutorials, and related blog posts? Are anchor texts descriptive (not "click here" or bare URLs)? Is there a clear relationship to a pillar page or content cluster, or does this piece feel orphaned?

• Link format: Are all links full URLs (starting with https://)? Relative paths break when content is handed to content marketers or published outside the CMS.
• External link completeness: Does every section discussing a competitor or third-party product link to that product's official page?
• Source links: Are factual claims (deprecation dates, release announcements, benchmark numbers, pricing) linked to their original sources?

What to output: Note which internal links are present and flag obvious gaps (e.g., a comparison piece that mentions Tiger Data's product but doesn't link to the relevant product or docs page). Flag generic anchor text. Flag any relative paths (not full URLs). Flag sections that mention competitor products without linking to their official pages. Flag unlinked factual claims as "needs source."

Comparison piece note: Links to Tiger Data's own product should feel organic — earned by context, not forced. A comparison piece that links to Tiger Data on every mention reads as promotional. One or two well-placed links to docs or a relevant case study is appropriate.

---

Dimension 6: Editorial Neutrality (comparison and roundup pieces only — skip for single-topic SEO articles)

What to evaluate: Are alternatives presented fairly? Are the evaluation criteria applied consistently across all options — not just to Tiger Data's strengths? Does the piece acknowledge Tiger Data's product limitations where relevant? Would a reader who has no prior opinion trust this as an objective resource, or does it read as marketing disguised as editorial?

• Competitive freshness: For each competitor mentioned, verify via web search that the product still exists as described, the pricing model hasn't changed, and no major rebranding or discontinuation has occurred since the article was written.

What to output: Assess whether evidence is symmetric. Flag asymmetric evidence — e.g., Tiger Data's performance backed by benchmarks, competitors described in vague generalities. Flag criteria that appear to be designed to favor one outcome. Note whether the piece's recommendation (if any) feels earned by the analysis or predetermined. Flag stale competitor information with the current source URL. Note which competitors were verified and which could not be checked (e.g., if web search is unavailable).

Why this matters: Asymmetric evidence is the core credibility failure for comparison content. Readers evaluating tools are skeptical — they've read vendor-written comparisons before. A piece that clearly advocates for one option from the structure undermines the editorial trust that makes comparison content rank and convert.

---

Dimension 7: Technical SEO Readiness

What to evaluate: Heading hierarchy (single H1, logical H2/H3 nesting — no skipped levels), paragraph length (short paragraphs, ideally under 4 sentences), subheading frequency (roughly every 200–300 words for scannability), image alt text (if images are referenced), and whether the implied URL slug is clean and keyword-inclusive.

What to output: Flag any heading hierarchy violations. Note if paragraphs are consistently too long for web reading. Call out missing alt text if images are present. If the title implies a slug that would be unwieldy or keyword-poor, flag it.