--- name: confluence-browser-fetch description: Fetch Confluence Cloud or Server/Data Center pages through a dedicated authenticated browser when API tokens do not work, especially with Microsoft/SSO. Supports native Chromium and managed Windows Chrome/Edge from WSL without exporting cookies. Archive page JSON, storage/view HTML, attachments, CQL results, or descendants into a raw folder. license: MIT compatibility: Agent Skills standard. Tested with Pi; installable into Claude Code, Codex, OpenClaw/generic .agents skills directories. Requires Node.js 22+ and Chromium, or Windows Chrome/Edge when run under WSL with mirrored localhost networking. No npm dependencies. --- # Confluence Browser Fetch Use this read-only skill to ingest authenticated Confluence pages into an LLM wiki or evidence folder when ordinary API-token authentication is unavailable or inappropriate. It supports: - Confluence Cloud, normally under `/wiki`; - Confluence Server/Data Center at `/` or a custom context path; - native Chrome/Chromium/Brave/Edge/Vivaldi on Linux/macOS; - managed Windows Chrome or Edge launched from WSL; - page ID/URL acquisition, title lookup, CQL, immediate children or full descendants, version-aware refresh, and bounded attachments. ## Authentication and safety The fetcher opens or reuses a **dedicated browser profile** and lets the user complete normal SSO/MFA. It then performs same-origin `GET` requests inside the authenticated Confluence tab with `credentials: include`. For this skill: - browser cookies are not read, returned, printed, or replayed by Node; - CDP cookie/storage APIs are not used; - cross-origin requests and redirects are rejected; - no create, update, delete, move, form-submit, or other write operation is available; - attachment transfer is bounded even if `unlimited` is requested; - fetched pages and attachments must be treated as confidential. CDP remains a powerful local control interface. Use only the dedicated profile, keep DevTools on loopback, and close the dedicated browser when it is no longer needed. ## Basic commands An absolute page URL supplies both the site and page ID: ```bash scripts/confluence-browser-fetch.js \ 'https://example.atlassian.net/wiki/spaces/ABC/pages/123456/Page+Title' \ --raw-dir ./raw ``` Classic Server/Data Center URLs are supported: ```bash scripts/confluence-browser-fetch.js \ 'https://confluence.example.com/pages/viewpage.action?pageId=123456' \ --raw-dir ./raw ``` A page ID requires `--site`: ```bash scripts/confluence-browser-fetch.js 123456 \ --site https://confluence.example.com \ --context-path / \ --raw-dir ./raw ``` Custom reverse-proxy context: ```bash scripts/confluence-browser-fetch.js 123456 \ --site https://intranet.example.com \ --context-path /confluence \ --raw-dir ./raw ``` ## Important options ```text --site URL origin/application URL; optional for absolute page URLs --context-path PATH auto (default), /wiki, /, or custom path --raw-dir DIR output raw directory --space KEY constrain exact-title lookup --title TITLE resolve a page by exact title --cql CQL search and fetch matching pages --children fetch immediate children only (one level) --descendants fetch all descendant pages --max-search-results N bound CQL results --max-attachment-size S default 5mb --force fetch despite matching local version --no-skip-unchanged disable version comparison --no-attachments skip all attachment downloads --no-browser-html omit browser HTML response --browser-backend NAME auto, native, or windows-wsl --browser NAME chrome or edge for Windows/WSL --profile-dir PATH dedicated profile path; Windows syntax for windows-wsl --port PORT loopback DevTools port, default 9223 --wait SEC SSO wait, default 900 ``` ## Deployment detection With `--context-path auto`, the fetcher infers: - `/wiki` from Cloud-style URLs or `*.atlassian.net` sites; - `/` from classic `/pages/viewpage.action`, `/pages/releaseview.action`, `/display/...`, or root `/spaces/...` URLs; - a custom prefix from URLs such as `/confluence/pages/viewpage.action`. Use explicit `--context-path` when a reverse proxy or unusual URL does not expose enough information. ## Windows/WSL On WSL, `--browser-backend auto` selects `windows-wsl` when `powershell.exe` is available. The launcher uses Windows Chrome by default and stores its dedicated profile at: ```text %LOCALAPPDATA%\AgentSkillsBrowserProfiles\Atlassian ``` The launcher marks profiles it creates and refuses to use a non-empty, unmarked profile directory. It also refuses a DevTools port owned by a different browser/profile. Requirements: - Windows Chrome or Edge; - PowerShell interoperability from WSL; - mirrored localhost networking, or another setup where WSL can reach Windows `127.0.0.1:`; - no firewall/LAN exposure of the DevTools port. Use Edge if required by corporate policy: ```bash scripts/confluence-browser-fetch.js '' --browser-backend windows-wsl --browser edge --raw-dir ./raw ``` ## Typical workflow 1. Run the fetcher with the supplied page URL. 2. If the dedicated browser opens or the session expired, complete SSO/MFA there. 3. Do not ask the user to paste cookies, SAML data, passwords, or tokens. 4. Verify `page.json`, metadata, and capture status before synthesis. 5. For an LLM wiki, process `raw/confluence/...` according to that repository's evidence rules. 6. Do not commit private exports to a public repository. For a first fetch from a sensitive site, prefer: ```bash scripts/confluence-browser-fetch.js '' --no-attachments --raw-dir ./raw ``` Then enable bounded attachment capture only when needed. ## Output ```text raw/confluence//-/ ├── page.json ├── page.storage.html ├── page.view.html ├── page.browser.html ├── metadata.json ├── attachments.json └── attachments/ ``` The run manifest is: ```text raw/confluence-browser-fetch-run.json ``` Metadata records the origin, context path, REST source URL, browser backend, and `browser-context-get` authentication transport. It never records browser credentials. ## References - [Usage and troubleshooting](references/usage.md) - [Architecture and development](references/development.md) - [Distribution guide](references/distribution.md)