SEO Brief Writer with an Editorial Standard

Turns a keyword into a complete content brief — outline, entities, FAQs — grounded in SERP data and archived to a brief library.

The dirty secret of content operations is that the brief is the bottleneck, not the writing. A good brief takes a strategist ninety minutes: pull the SERP, read ten competing pages, extract the heading themes, find the gap nobody covers, list the entities that make the piece credible, collect the people-also-ask questions. Skip any of it and the writer either asks you eleven questions or — worse — doesn't, and you get 1,800 fluent words that rank for nothing.

So teams ration briefs. The calendar says twelve posts a month; the strategist produces five real briefs and seven "just cover the basics" Slack messages. The five perform. The seven don't. Everyone concludes content doesn't work.

seo-brief-writer makes the brief the cheap part again — and it does it with an auditable editorial standard, not a lucky prompt.

The standard is a skill

The entire house format lives in one skill in agent.js — six numbered sections, in order, every time:

const briefFormat = skill("brief-format", {
  description: "The house structure for a content brief: title options, outline, entities, FAQs, and internal links, grounded in SERP data.",
  instructions: "Brief structure - every section, in this order:\n1. Target: the keyword, inferred search intent...\n3. Outline: H2s and nested H3s that cover every heading theme the top results share, plus at least one angle none of them cover - mark it GAP.\n...",
  tools: [serpLookup, tools.files],
});

Notice section 3's rule: cover every shared heading theme and add at least one angle none of the competitors have, marked GAP. That's the difference between a brief and a summary of page one — and because it's a skill, it's versioned. When your SEO lead decides briefs need a schema-markup section, that's a one-line edit and a redeploy, applied identically to every brief afterward. While the skill is active, the agent's tools narrow to exactly serp_lookup and files: research and write, nothing else.

Grounded, by contract

The identity draws a hard line most "AI SEO tools" blur:

"serp_lookup is your only source of search-landscape data; call it exactly once per brief and ground every competitive claim in what it returns. Never invent rankings or search volumes."

serp_lookup is the file's one stub — a custom tool whose run body you point at DataForSEO, SerpAPI, or your own crawl. The sandbox has no network, so this tool is the only window to the SERP, which makes the grounding rule enforceable in the trace: every competitive claim in a brief maps to one visible tool call.

A library, so briefs link to briefs

Every brief lands at briefs/<keyword-slug>.md on the agent's durable filesystem — and section 6 of the format uses that: internal-link suggestions reference other briefs in the library. The more you brief, the more interlinked your content plan becomes, automatically. The library task reads the index for zero tokens:

agent.task("library", async () => {
  const names = await agent.files.list("briefs");
  return Array.isArray(names) ? names.sort() : [];
});

Each brief run is one managed loop capped at { maxSteps: 12, maxCost: 0.6 }, under a $3/day identity budget — a full month's calendar of briefs for less than one strategist-hour.

Deploy it

git clone https://github.com/oncell/oncell-cookbook && cd oncell-cookbook
cp .env.example .env   # add your ONCELL_API_KEY
npm run deploy seo-brief-writer

curl -X POST https://api.oncell.ai/api/v1/agents/seo-brief-writer/brief \
  -H "Authorization: Bearer $ONCELL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"keyword": "warehouse slotting optimization"}'

The answer is deliberately terse — path, recommended title, H2 list — because the deliverable is the file. Open briefs/warehouse-slotting-optimization.md and hand it to a writer; rerun the same keyword next quarter and the identity's refresh rule overwrites it with the current SERP landscape.

What you didn't have to build

A SERP-scraping pipeline with storage. A templating system for brief structure. A shared drive taxonomy that three people maintain and nobody follows. Version control for your editorial standards — the skill diff is the changelog. Spend controls for the day someone briefs 400 keywords — the per-run cap and daily ceiling are runtime-enforced.

You wrote your editorial standard once and pointed one stub at your SERP provider. Brief production is now a curl in a loop — and the writer never has to ask what you meant.