Brief Outline Generator

Generates a content outline as a formatted .docx file. The output is a skeleton — section headings, short topic prompts, angles for each section — that a writer fills in with their own conclusions, numbers, and prose.

This is an outline generator, not a brief generator. If your output reads like an article in note form, you've gone too far. Read references/section-rules.md before generating anything.

The DOCX is always produced by running scripts/generate-brief.py. Do not reimplement the renderer. Do not write inline docx code. Assemble the config JSON and run the script.

Inputs — collect from user before proceeding

Execution workflow — follow these steps in order

Step 1 — Validate inputs

• title: non-empty; warn (don't block) if > 70 chars
• focus_keyword: non-empty
• sitemap_url: non-empty, starts with http:// or https:// (accept any path — .xml, .xml.gz, or extensionless dynamic URLs are all valid)
• word_count_range: two positive integers separated by - (e.g. 1500-2000)
• target_intent: one of Informational, Commercial, Transactional, Navigational (case-insensitive)

Derive two values from sitemap_url:

• domain_url — strip the path to get the origin, e.g. https://firefly.ai/sitemap.xml → https://firefly.ai. Used in the config JSON.
• domain_hostname — strip the protocol too, e.g. https://firefly.ai → firefly.ai. Used in tool calls that require a bare hostname (e.g. site:firefly.ai).

Stop and report all validation errors before proceeding.

Step 2 — Read the rules

Read references/section-rules.md in full now. It contains:

• The outline-vs-brief distinction (the most important rule)
• Hard bullet rules (≤ 12 words, no invented numbers, no conclusions, no em-dash clauses)
• Hard structure rules (no topic_summary, no Writer Directives box, TLDR carries 2–3 topic pointers)
• The four archetypes (Listicle, Comparison, How-to, Concept/Explainer) and their section sets
• Per-section rules and good/bad bullet examples
• A final quality check to run before generating the DOCX

Do not skip this step. Generating without reading the rules produces brief-style output every time.

Step 3 — Domain analysis

DataForSEO tools are deferred — load them before calling. Call tool_search(query="on_page content parsing") at the start of this step.

3a — Discover site URLs

Call web_fetch on sitemap_url.

• Readable XML (response contains <loc> tags) → extract every <loc> URL. These are your site URLs.
• Binary / compressed (unreadable response) → fall back: call dataforseo:serp_organic_live_advanced with keyword="site:{domain_hostname}" (e.g. site:firefly.ai) to get all indexed URLs.

Cap at 20 URLs. If more exist, prioritise: homepage → product/use-case pages → blog/resource pages.

3b — Read full page content

For each URL from 3a, call dataforseo:on_page_content_parsing with enable_javascript: true.

This returns fully rendered page text — headings, body copy, product descriptions, etc. Collect all output.

If a page fails (bot protection, timeout), skip it and continue — do not abort.

3c — Extract meta title and description

From the homepage content parsed in 3b, extract:

• Page <title> → meta_title
• <meta name="description"> content → meta_description

Surface character count advisories if outside recommended ranges:

• meta_title outside 50–60 chars → "Note: fetched meta title is N chars (recommended: 50–60)."
• meta_description outside 150–160 chars → "Note: fetched meta description is N chars (recommended: 150–160)."

These are passed into the config as meta_title and meta_description and rendered in the metadata table.

3d — Compile domain_context

From all parsed page content, extract and store:

• Product name and core value proposition
• Key technical terms and vocabulary used on the site
• Target audience signals (roles, team types, use cases mentioned)
• Existing content topics — used to avoid duplication in the outline

Store as domain_context. Do not render domain_context in the output document. If all fetches fail, set domain_context = null and continue.

Step 4 — Classify the title's archetype

Based on the title pattern, pick one of:

• Listicle / Tool Roundup — "Top N", "Best X", "X Alternatives"
• Comparison / Versus — "X vs Y", "X or Y"
• How-to / Implementation Guide — "How to X", "How do X teams Y", "Implementing X"
• Concept / Explainer — "What is X", "Designing X", "[Function/Pattern]: Designing X"

If multiple seem to fit, use the defaults from section-rules.md. If still unclear, ask the user.

Announce the chosen archetype to the user before generating — one line, e.g. "Detected archetype: How-to. Generating outline with intro → strategies → case studies → implementation → FAQs." Let them override if they disagree.

Step 5 — Generate keyword volumes

Fetch USA monthly search volumes for the focus keyword and each secondary keyword using DataForSEO.

DataForSEO tools are deferred — load them before calling.

1. Call tool_search(query="keyword search volume google ads") to load the DataForSEO keyword tools.
2. Call dataforseo:kw_data_google_ads_search_volume with the full list of keywords (focus + all secondaries) in a single call. Use location_code for the USA (2840).
3. From the response, extract the search_volume field for each keyword.
4. Format each volume as a thousands-separated string (e.g. 3400 → "3,400"). Volumes under 1,000 stay as plain digits (e.g. "500", "30"). Volume of 0 should be rendered as "0", not "N/A" — it's a real datapoint.
5. If a keyword returns no data, set its volume to "N/A" and continue. Don't abort the whole run.
6. If tool_search returns no DataForSEO tools (connector not installed), set every volume to "N/A" and surface a one-line warning: "DataForSEO connector not available — keyword volumes set to N/A."

Never omit the volume field on any keyword. Every row must have one.

Flag volume mismatches. If the focus keyword's volume is more than 10× smaller than any secondary keyword's volume, tell the user before generating: "Note: your focus keyword has volume X, but secondary keyword Y has volume Z. Consider whether Y should be the focus." Let them decide; don't auto-swap.

Store as:

"focus_keyword_volume": "2,400",
"secondary_keywords": [
  { "keyword": "disaster recovery plan", "volume": "1,900" },
  { "keyword": "cloud DR strategy",      "volume": "N/A"   }
]

If no secondary keywords were supplied, generate 5 by combining base terms from the focus keyword, top domain key terms, and modifiers ("best practices", "guide", "checklist", "for teams", current year). Then fetch volumes for them the same way.

Step 6 — Build the outline using the archetype's section set

Use the section set for the archetype you chose in Step 4. Do not force every topic into a how-to template. A listicle has no Problem or Case Studies section. A comparison has no Implementation steps.

Each section object shape:

{
  "heading": "H2",
  "title": "Section Title",
  "rules": ["short topic prompt 1", "short topic prompt 2"],
  "subsections": [
    {
      "heading": "H3",
      "title": "Subsection Titl