<aside> ✍️
This stage writes from an approved brief rather than from a topic alone. Fresh research is added as supporting context, but the approved brief remains the route the writer must follow.
</aside>
flowchart LR
A["Approved brief"] --> B["Download + parse brief"]
B --> C["AI: Find 3 new sources"]
C --> D["Fetch article content"]
D --> E["AI: Deconstruct sources"]
E --> F["AI: Write article"]
F --> G["Markdown → HTML"]
G --> H["Create Google Doc"]
H --> I["Final Articles folder"]
I --> J["Human review"]
Trigger On New Brief watches the approved-brief folder. Extract Brief Details separates title, size, meta, recommended keywords and article structure from the document.Find 3 relevant articles searches for three deliberately different source types: a statistics source, a real-world case study and a technical-background article. It must verify accessibility and return empty values rather than invent missing links.Fetch the structure of each of 3 articles, which extracts the usable factual material.Write the article receives the approved brief, article structure, keywords and all three source packs. Its system/user prompt pair controls voice, grounding, evidence density, article structure, repetition, uniqueness checking and final JSON output.Approval happens before the expensive writing stage: moving the brief into the approved folder. A person still reviews the finished article before publication. The JSON also contains a later Webflow export branch: once an article is moved into Approved articles, the system parses its slug, title, meta, H1, locale and image, creates the correct RU/EN/KK Webflow item, and notifies the editor. Publishing remains a human action.
| Node | Why it matters |
|---|---|
Extract Brief Details |
Converts the approved Google Doc into explicit writing inputs instead of handing an undifferentiated document to the writer. |
Find 3 relevant articles |
Research agent with source-type diversity and a strict verified-link output contract. |
Process Article Structures for Writing Agent |
Normalizes fetched page content into three stable text fields before the final writing agent receives it. |
Write the article |
Main writing agent. It also has a uniqueness-check tool, memory and a structured output parser. |
Post in Google Docs |
Creates the final Google Doc from generated HTML and applies document-level styling. |
Extract Docs Details |
In the export branch, extracts slug, title, meta description, H1, locale, article body and thumbnail before Webflow routing. |
<aside> 🤖
The exact prompts for source discovery, source deconstruction and final article writing are in Prompts & Code Guide.
</aside>