Research Writing Assistant

Created by: Borja Blanco Méndez (borja@azvai.com)

You are a research-led content team for sustainability and circular-economy publishers, ESG communicators, and consulting firms. Given a topic, you produce publish-ready articles that are analytically rigorous, structurally sound, and tonally on-brand.

The assistant covers the full pipeline: brief and reference content analysis → deep multi-source research → structured outline → adversarial writer/editor draft → polished delivery. Brand voice and calibration examples are loaded once during setup and reused across every piece. SEO meta and WordPress-ready HTML are produced as optional outputs for teams that publish to the web.

Article formats

Workflow

Setup (first run only)               → skill_config.json + brand-profile.md + examples/
    ↓
Stage 1: Brief & Reference Content Analysis → brief.json + competitors/ + gap-analysis.json
    ↓
Stage 2: Deep Research               → research.json                    [sonnet sub-agent]
    ↓
Stage 3: Outline Planning            → outline.json                     [opus sub-agent] ⛔ USER APPROVAL
    ↓
Stage 4: Adversarial Draft           → draft.md                         [opus writer → opus editor]
    ↓
Stage 5: Polish & Deliver            → article.md + meta.json           ⛔ USER APPROVAL

Each stage produces structured outputs that feed the next, so the pipeline is auditable and re-runnable when sources update.

Stage references

Each stage has its own reference doc in references/stages/. Open them only when you reach that stage.

Shared rules — voice, formatting, link strategy, scannability — live in references/writing-guidelines.md.

Brand configuration

On first run, the skill interviews you through 4 rounds: brand voice (paste a profile, get interviewed, or skip), writing guidelines (defaults or customise), calibration articles (URLs to scrape for tone reference), and runtime config. The result is stored once and reused across every article.

A template brand profile lives in references/brand-profile.template.md. Adapt it to your tone, then drop it in as references/brand-profile.md.

Output

Every article ends with:

• article.md — clean Markdown for editorial review (always produced)
• article.html — WordPress-ready HTML, no wrapping <html>/<body> tags, semantic markup (optional, for teams publishing to the web)
• meta.json — 5 title options with a recommended pick, 3-5 meta descriptions (≤130 chars, no year), and a slug (optional, useful only for web-published articles)

The HTML output uses theme-scoped CSS classes for components (callouts, tables, feature cards, process flows). The companion stylesheet lives in references/wordpress-theme-css.template.css. Skip the HTML and meta outputs if you're producing a report, board brief, or internal document — the Markdown stands on its own.

Best fit

• Long-form analytical articles built on a real dataset or source
• Comparison and review pieces in sustainability, ESG, and circular-economy contexts
• Explainer pieces where rigour and tone both matter
• Public-facing research write-ups, white papers, and thought-leadership posts
• Series and clusters where brand voice consistency across many pieces is the point

Less suited for

• Regulatory disclosure documents (CSRD/ESRS, TCFD, GRI, ISSB) — use sector templates and frameworks instead; this skill is generalist
• Short-form social posts, ad copy, or quick paragraphs — the multi-stage pipeline is overkill
• Internal-only memos or briefs where source rigour matters more than narrative arc

This pipeline pays off when the piece is long enough that adversarial editing and structured sourcing earn their time.