Content Brief Factory

Content Brief Factory

A composite over site-content-catalog, robomotion-serp Search/Auto-Complete, and customer-voice mining. You (the agent) synthesize each brief; two bundled keyless scripts do the deterministic glue (coverage match + competing-page stats).

When to use

• "Create content briefs for these keywords / our Q2 calendar / our pSEO templates."
• "Generate 20 briefs from this keyword list" / "what should we write about [topic]?"
• Downstream of site-content-catalog (coverage) and any keyword-discovery skill.

How to run

1 — Existing-coverage check

Catalog the site, then match keywords against it:

# (a) inventory the site
python3 ${SKILL_DIR}/../site-content-catalog/scripts/site_catalog.py \
  --domain example.com --deep 0 --output ${WORKSPACE}/catalog.json

# (b) normalize keywords + flag new vs update + internal-link candidates
python3 ${SKILL_DIR}/scripts/load_keywords.py \
  --input ${WORKSPACE}/keywords.csv --catalog ${WORKSPACE}/catalog.json \
  --output ${WORKSPACE}/coverage.json

2 — SERP analysis (agent, via serp)

Per keyword, run robomotion-serp Search (localized, top-10) → ranked {title, description, url}, and Auto Complete to mine PAA-style related queries. This is the keyless default. If DATAFORSEO_LOGIN/DATAFORSEO_PASSWORD (or SERPER_API_KEY) is set → pull the SERP from that paid API instead for richer/cleaner top-10 results; if not set → keyless robomotion-serp (the default). Volume/difficulty have no free source — degrade to serp-derived signals and label numbers as estimates.

3 — Competing-page analysis

Feed the top SERP URLs to the page-stats helper:

python3 ${SKILL_DIR}/scripts/fetch_pages.py \
  --input ${WORKSPACE}/serp_top10.json --max 5 --output ${WORKSPACE}/pages.json

Emits per-page {title, content_type, word_count, headings, image_count, list_count, link_count, published_hint} + avg_word_count. For anti-bot pages, the agent escalates to web-automation + Robomotion Proxy.

4 — Customer-language mining (agent)

Reddit + reviews for the topic → real questions, pain language, misconceptions, quotable proof. Lighter path: robomotion-serp Search site:reddit.com <topic> + Extract Content. At scale / hostile sites: Apify review/Reddit actor (APIFY_API_TOKEN).

5 — Synthesize each brief (you)

Fold landscape + gap + customer quotes + ICP + brand voice into: metadata, search-landscape table, SERP-feature targets, differentiation angle with quotes, recommended outline, internal-linking plan (from coverage.json), content requirements, competitive-advantage checklist. For covered keywords, emit an update brief, not a new-page brief.

6 — Batch output

One markdown brief per keyword + a ranked summary table + CSV. Persist to Airtable/Supabase for the editorial calendar if asked.

--help is available on both bundled scripts.

Outputs

One markdown brief per keyword; batch runs add a ranked summary table + CSV. Saved to workspace / Agent Teams channel.

Credentials / env

• Required: none — serp Search/Auto-Complete + the bundled keyless scripts cover the core; the agent is the synthesis engine.
• Optional:
  • DATAFORSEO_LOGIN/DATAFORSEO_PASSWORD (or SERPER_API_KEY) — paid SERP for step 2. If set → cleaner/richer top-10 SERP. If not → keyless robomotion-serp Search + Auto-Complete (the default).
  • APIFY_API_TOKEN — Reddit + review-site scraping at scale (hostile-site fallback). If set → at-scale voice mining. If not → degrade to serp site: searches + web-automation with thinner voice mining (the default).
  • ANTHROPIC_API_KEY / OPENAI_API_KEY — not used by any script (the agent is the model).
  • A paid SEO-metrics API (Semrush/Ahrefs/DataForSEO) for volume/difficulty — no Robomotion package; out-of-band optional enrichment only. Absent → label metrics as serp-derived estimates (the default).

Notes & edge cases

• A great brief states what ranks, why it wins, the gap, the customer language proving the gap matters, and how to structure a page that beats it.
• Run keywords in batch; cap concurrency and route through Robomotion Proxy to avoid blocks.
• Volume/difficulty are best-effort without a paid API — label them estimates.