| name | obscura-web-browse |
| description | Browse and extract web pages with Dockerized Obscura. Use to open URLs, fetch JS-rendered pages, extract text/links/assets, run simple page scripts, or inspect page resources. |
Obscura Web Browse
Use Obscura through Docker to browse known URLs and return useful page content or instrumentation. Prefer the smallest path: one-off docker run fetch first, detached CDP only when the task needs stateful interaction or full network events.
Quick Start
- Read Obscura Docker Runbook for Docker launch patterns.
- Read Browse Workflows for the extraction path to use.
- Read Instrumentation And Limits before promising network data, screenshots, PDFs, local access, or stealth behavior.
- Use CDP Workflows only when the task needs stateful browsing, scroll/lazy-load interaction, or full Network events.
- Use
scripts/obscura_browse.py when a repeatable Docker fetch is useful.
Workflow
- Confirm the user gave a known URL; if they gave a query instead, use a web-search skill.
- Choose output:
markdown or text for reading,links for navigation,assets or --include-assets --json for resource inspection,html for parser/debug fallback,--eval for targeted DOM/state extraction.
- Run Dockerized Obscura with explicit wait/timeout choices.
- If output is empty or only an SPA shell, use the fallback in Browse Workflows: inspect HTML, dump assets, fetch likely JS bundles with
--dump original, then extract embedded docs/data if present.
- Summarize or extract from fetched output; follow only the links needed to answer.
- Report command failures, empty output, blocked pages, and unsupported visual/PDF requests plainly.
Sample Applications
See Sample Applications for browse-and-summarize, dynamic-section extraction, resource inspection, and limitation examples.
Reference Map
