Skip to content

limitations.md

Latest commit

 

History

History
34 lines (19 loc) · 2.35 KB

limitations.md

File metadata and controls

34 lines (19 loc) · 2.35 KB

Limitations

Where ClawReflex works best

  • Repetitive browser workflows on stable sites
  • Sites with accessible ARIA roles, labels, and semantic HTML (locators stay stable)
  • Workflows with clear, file-based output (CSV downloads, report exports)
  • Tasks that run on a fixed schedule with predictable inputs

Where it is weaker

Anti-bot surfaces. Sites with aggressive bot detection (Cloudflare Turnstile, PerimeterX, DataDome) will block Playwright. No workaround exists for these without a real authenticated browser session.

Repeated MFA. Workflows requiring a new MFA code on every run cannot be automated unattended. The workaround is to save a storage state after a manual login and reuse it until the session expires — see troubleshooting.md.

Frequent redesigns. Sites that overhaul their UI often will require frequent repair sessions. The deterministic macro is only as stable as the site it runs on.

Tasks that are mostly judgment. If the right action depends on reading and interpreting content (e.g. "find the invoice with the largest overdue balance and download it"), the macro lane cannot handle it. These require a model on every run, which is a different architecture.

Dynamic DOM without accessible locators. Sites that use auto-generated class names or position-based selectors produce fragile locators that break on every deploy. Use data-testid attributes where possible.

Multi-tab workflows. The emitter does not support workflows that require opening or interacting with multiple browser tabs. Single-page, single-tab flows only.

File uploads. The current action set does not include a fileUpload action. Workflows that need to upload a file to a site are not supported without a custom emitter extension — see expert-extension-guide.md.

Infinite scrolling / pagination. The emitter has no built-in loop construct. Workflows that need to scroll through multiple pages of results and collect them are not supported without a custom step.

This is not a general browser agent

ClawReflex is purpose-built for repetitive, predictable, file-producing workflows. It is not a general-purpose browser agent that can handle open-ended tasks, make decisions, or adapt to unexpected situations mid-run. For those use cases, use OpenClaw directly with the browser tool enabled.