Vector Browser Automation

What It Solves

Browser tasks often break when page structure changes, selectors drift, or there is no evidence for what happened. Copying page text or relying on pixel-based clicking can fail after minor UI changes and makes every step hard to audit. Vector constrains browser work into an observable loop: verify the environment, then run RECON → PLAN → EXECUTE → COLLECT → REPORT, and return status, logs, screenshots, data, and error summaries.

How It Works

It is built around Playwright MCP and CDP, preferring accessibility snapshots over purely visual recognition. Common paths include:
- Data collection: extract structured content with getByRole, getByLabel, or data-testid, then save JSON/CSV.
- Forms and evidence: capture screenshots before and after fill, submit, or upload actions, and record console/network errors.
- Network analysis: capture HAR files, request/response details, and console logs through CDP.
- Special cases: mobile emulation, parallel sessions, and visual fallback for shadow DOM or canvas elements.

Boundaries

It fits browser-bound evidence collection, scraping, forms, and bug reproduction. Public APIs should be preferred when available. It does not encourage bypassing CAPTCHAs, paywalls, or ToS, and it should not run sensitive work against shared production credential profiles.