Content Brief Builder

Give it a keyword and what's ranking; get back a brief a junior writer can execute into a senior article. The brief makes every strategic decision once — intent, angle, structure, coverage, length, links, SEO targets — so the writer spends their energy on prose, not on re-litigating strategy mid-draft. A great brief is the difference between an article that ranks and an article that merely exists.

This skill plans; it does not draft. It will not write the body copy, and it will not invent the SERP — if no real SERP/research data is provided, it says so and offers to run the upstream research first, rather than guessing what's ranking.

Skills this calls

• brand-brain (required) — resolves and loads the active brand's voice, banned words, ICP + awareness tendency, offer/destinations, positioning, and real proof. The brief steers the writer toward the brand's real story; it does not re-derive brand context here.
• (optional, when installed / when the brief naturally needs them)
  • serp-research / SERP Analysis Report — if no SERP data was provided, this is the upstream step; call it (or ask the user to) before briefing.
  • topic-cluster-pillar-architect — to slot this article into its cluster and pull the right internal-link targets (pillar ↔ spokes).
  • proof-vault — to pull real stats/quotes/case studies the article should cite (so the brief names which proof goes where).
  • cta-variant-generator — to specify the closing CTA in the brand's voice and offer.
  • positioning-messaging-architect / icp-persona-builder — already folded into brand.md via brand-brain; call directly only for deeper pillars/persona language.

Synthesize a section inline only when a needed component is unavailable.

How a run works

Step 0  Load the brand   ──► call the `brand-brain` skill (it bootstraps on first use)
Step 1  Gather inputs    ──► keyword + SERP/research data (or trigger upstream research)
Step 2  Build the brief  ──► run the IDEALS checklist (our working model), section by section
Step 3  Self-review against the checklist, then present and offer to persist

Step 0 — Load the brand (always first)

Invoke the brand-brain skill (Skill tool, skill: brand-brain), passing the request and any named brand. It returns the digest — voice adjectives, banned words, offer mechanics + destination URLs, real proof, positioning line, ICP + awareness tendency — and the brand.md path. The brief obeys voice + banned-words as hard overrides, points the writer only at real proof (anything else marked [verify]), and frames the angle through the brand's positioning. Do not build the brief until brand-brain returns.

Fallback if brand-brain is not installed: read ~/.brandbrain/brands/.active + that brand's brand.md directly; if none exists, ask the user to install brand-brain or answer a 4-question mini-setup (what it is · ICP + awareness · offer + destination · 3 voice adjectives + banned words), then proceed.

Step 1 — Gather the inputs

You need two things before you can brief: the target keyword and real SERP/research data (the top-ranking pages' structure, formats, gaps, and common patterns). If the user hands over a SERP research dossier, use it. If they hand over only a keyword, do not invent the SERP — offer to run serp-research (or ask them to paste the SERP) first. Also capture, if available: secondary keywords, the cluster/pillar this belongs to, and any deadline or length constraint.

IDEALS — our working brief checklist

IDEALS is our own organizing mnemonic for this skill, not an established industry framework. Every brief answers six questions, in order. Earlier answers constrain later ones — get intent wrong and the whole brief is wrong.

I — Intent (decide this first)

Classify the keyword's dominant intent from the SERP, not from the words alone — the SERP is the ground truth for intent. If the top ten are listicles, Google has decided this is informational-list; a 3,000-word ultimate guide will not rank there.

Pin the awareness stage (from the brand's ICP tendency + the keyword) and the JTBD — what the reader is hiring this article to do. These set the CTA ceiling later (don't ask a problem-aware reader to buy).

D — Differentiated angle (the part juniors skip)

A brief that just re-skins the SERP produces a page that ranks #11. Use the research dossier's gaps and common patterns to choose a wedge:

• Cover what every result omits — the gap they all share (a step, an objection, a real number, a use case).
• Beat them on format — original data, a real example, a tool/template, a decision table, expert POV — where the SERP is all thin rehash.
• Frame through the brand's positioning — the angle should let the brand's real differentiator and proof show up naturally, not as a bolt-on pitch. Mark the moment in the outline where the brand's POV earns its place.

Write the angle as one sentence the writer can hold in their head: "Unlike the listicles, this is the only guide that [wedge], proven with [real proof]." If you can't find a genuine wedge, say the topic may not be worth writing — don't manufacture a fake differentiator.

E — Entities & coverage (topical completeness)

List the entities, subtopics, and terms the page must mention to be recognized as comprehensive (pull from the SERP's shared vocabulary, related searches, and the brand's domain — not a generic LSI dump). Then list the must-answer questions (People-Also-Ask + the real questions the ICP asks). Coverage is a floor, not a keyword-stuffing quota: every entity must earn a place in the outline.

A — Architecture (the outline)

Build a heading-by-heading outline (H1 → H2 → H3) where every section has a job:

## [H2 working heading]
   Intent of this section: [what the reader gets here]
   Must cover: [entities/questions assigned to this section]
   Proof/asset: [stat, quote, example, or "[verify]"]  ·  Internal link: [anchor → target]
   Target length: [~words]

Rules: lead with an answer-first opening for informational/featured-snippet plays (a 40–55-word direct answer the page can win the snippet with); order sections by searcher logic, not by what's easy to write; assign each must-answer question to exactly one section; note where a table, list, or image belongs. Map the closing CTA section to the brand's offer and awareness ceiling.

L — Links & assets

• Internal links: specific anchor text → specific destination URL, pulled from the cluster (call topic-cluster-pillar-architect if present). Always include the path up to the pillar and across to sibling spokes. Note 2–5 high-value internal links minimum.
• Proof: name the real stat/quote/case study per section (call proof-vault); anything unconfirmed is [verify] — never a placeholder number presented as fact.
• CTA: specify the closing call to action (call cta-variant-generator) matched to the article's intent and the reader's awareness stage.

S — SEO targets

Spell out the exact specs so the writer (or the On-Page SEO skill) has no guesswork:

• Primary keyword + 2–4 secondary/semantic variants and where each lands.
• SEO title (≤~60 chars, keyword near the front, on-voice).
• Meta description (≤~155 chars, promises the payoff + a reason to click).
• H1 (can differ from the title; reads naturally).
• URL slug (short, keyword, lowercase-hyphenated).
• Word count — derived from the SERP's ranking range, not a wish. State it as a target range with the rationale ("top results run 1,800–2,400; target ~2,200 to match depth without padding").
• Schema/format flags if relevant (how-to, FAQ, comparison table for snippet).

Output shape

# Content Brief — [keyword]   ·   Brand: [slug, via brand-brain]
Intent: [type · awareness stage · JTBD]
Angle: [one-sentence differentiated angle]
Word count: [range + rationale]  ·  Primary KW: […]  ·  Secondary: […]

## SEO targets
SEO title · Meta · H1 · URL slug · schema/format flags

## Entities & must-answer questions
[coverage list]  ·  [PAA / ICP questions]

## Outline (H1 → H2 → H3, with intent + coverage + proof + link + length per section)
…

## Internal links
[anchor → URL]  ×N

## Proof to cite   ·   ## Closing CTA
[real proof per section, [verify] flags]   ·   [CTA + destination]

## Writer notes
voice reminders · banned words · what to avoid · open [verify] items

Keep it tight enough to act on and complete enough to never need a second conversation.

Persistence

If the user wants it saved, write the brief to ./briefs/[slug]-[keyword].md (create ./briefs/ if needed) — never inside the skill folder, and never touch brand.md. Offer it; don't assume. A saved brief is the artifact the drafting engine reads next.

Principles

• Brand-brain first. No brief before brand-brain returns; its voice + banned-words override everything here.
• The SERP decides intent and length. Match the content type and depth Google is already rewarding; don't fight the SERP with a format it isn't ranking.
• Every brief needs a wedge. A re-skin of the top ten is a #11 page. If there's no honest differentiator, say so.
• Plan, don't draft. This skill makes strategic decisions and hands them off; it does not write the article.
• Coverage is a floor, not a stuffing quota. Entities earn their place in the outline or they're cut.
• Truth only. Real proof, real numbers, or [verify]. Never a fabricated stat or differentiator in a brief that a writer will treat as fact.
• One decision, one place. Every choice the writer would otherwise re-make mid-draft is made here, once.

What not to do

• Don't build a brief before brand-brain returns the active brand.
• Don't invent the SERP — if there's no real research, run/request serp-research first.
• Don't reimplement brand resolution/scanning/interviewing — call brand-brain.
• Don't write the article body, headlines as final copy, or the meta as deathless prose — this is a plan.
• Don't dump generic LSI keywords or pad word count; don't bolt the brand pitch onto an informational page.
• Don't fabricate proof, customers, stats, or differentiators; don't present [verify] items as confirmed.

Quality checklist (self-review before presenting)

• brand-brain called and the active brand loaded (or bootstrapped) before anything else?
• Real SERP/research data used (not invented); intent + length derived from the SERP?
• Intent, awareness stage, and JTBD pinned — and the CTA ceiling matches the awareness?
• A genuine one-sentence differentiated angle stated, framed through the brand's real positioning?
• Outline assigns every must-answer question and every entity to exactly one section, with intent + length per section?
• Internal links (anchor → real URL) and proof named per section, with [verify] on anything unconfirmed?
• SEO targets complete: title, meta, H1, slug, primary/secondary KWs, word-count range + rationale?
• Voice + banned-words honored; persistence to ./briefs/ offered (never the skill folder, never brand.md)?