Lyle is building a B2B analytics platform called Delphi. He has two engineers, a designer, and 67 pages of research — competitive analyses, customer interview notes, technical architecture docs, onboarding playbooks — scattered across Google Docs, Notion, and a shared Dropbox folder with no naming convention. Lyle's first enterprise prospect wants a security questionnaire completed by Thursday. The answers are in the research, but nobody can find anything. Lyle tells Reed to migrate everything into Confluence, organize it into a searchable knowledge base, and pull the security answers from the existing docs. Lyle has a product demo in 90 minutes and a pitch deck due tonight. The knowledge base cannot be his problem. ```typescript import { Page, Collection } from '@headlessly/content' const research = await Collection.create({ name: 'Research', description: 'Competitive analysis and market research' }) await Page.create({ title: 'Competitive Landscape Q1 2026', body: '## Market Position\n\nThree entrants since October...', collection: research.$id }) const security = await Page.find({ tags: ['security', 'compliance'] }) ``` ## Reed Opens the Confluence API Reed is Lyle's research agent. He is methodical — he reads documentation before writing a single request. Lyle gave him admin access to the Confluence Cloud instance at `delphi-analytics.atlassian.net`. Reed navigates to `developer.atlassian.com`. Authentication is the first obstacle. Confluence Cloud's API supports two methods: API tokens (basic auth with the user's email and a generated token) and OAuth 2.0 (three-legged, requiring a registered Atlassian app). The API token approach is simpler. Reed constructs the header: ```bash GET https://delphi-analytics.atlassian.net/wiki/rest/api/content?spaceKey=RES&type=page Authorization: Basic {base64(lyle@delphi-analytics.co:api_token)} ``` The response arrives. But it is not what Reed expected. ## The Atlassian Document Format Reed creates the first page — the competitive landscape analysis. He has the content as markdown. He sends: ```bash POST https://delphi-analytics.atlassian.net/wiki/api/v2/pages Content-Type: application/json { "spaceId": "65537", "status": "current", "title": "Competitive Landscape Q1 2026", "body": { "representation": "storage", "value": "## Market Position\n\nThree entrants since October..." } } ``` The page creates. Reed opens it in the browser. The content renders as a single block of plain text. The `##` is literal characters, not a heading. Confluence's storage format is not markdown. It is XHTML — a strict XML dialect with Confluence-specific macros: ```xml

Market Position

Three entrants since October. Two are VC-backed with $12M+ in funding. The third is a bootstrapped team that shipped faster than expected.

Last updated: January 2026

``` The `ac:structured-macro` namespace is Confluence's proprietary extension to XHTML. Tables, code blocks, status labels, expand sections, table of contents — all use `ac:` macros with their own attribute schemas. A simple expandable section requires: ```xml Security Certifications

SOC 2 Type II - In progress (est. Q3 2026)

ISO 27001 - Not started

``` Reed rewrites all 67 documents from markdown to Confluence XHTML. Each document averages 14 structural elements (headings, lists, tables, code blocks). Each element requires its Confluence XHTML equivalent. A markdown table becomes a `` with ``, ``, `

`, and `	` tags. A code block becomes `typescript`. A callout box becomes a `` or `