AI GitHub
Skyvern for running browser workflows
Build repeatable browser workflows around forms, navigation and extracted data.
Key Takeaways
- Skyvern combines a browser execution environment with model-guided steps and workflow definitions.
- A workflow needs a reachable test site and inputs safe to submit.
- Use a sandbox form with an optional field and a validation error.
- A browser action may have committed even if the agent missed the confirmation screen.
How it works
Skyvern combines a browser execution environment with model-guided steps and workflow definitions. Input parameters let the same workflow run with different records.
What you need
A workflow needs a reachable test site and inputs safe to submit. Store login and other secrets separately from the text you pass to model-driven steps.
What to try first
Use a sandbox form with an optional field and a validation error. Verify the saved record and retry behavior before adding a real destination.
Set an alternate viewport width and rerun the same form. A collapsed navigation menu or sticky control can change which interaction works.
Where to look for failures
A browser action may have committed even if the agent missed the confirmation screen. Design workflows around external state checks before retrying a write.
Related Franklin pages
Run multi-step agents on disposable tasks with known completion criteria. Inspect the actual side effects, especially when a tool times out after partially completing its work.
- Give a coding agent a workspace where it can edit files and execute development commands. OpenHands for software tasks in an isolated runtime
- Provide browser infrastructure that automation clients can connect to. Steel Browser for remote browser sessions
For an agent interface that reaches mobile apps, see Franklin’s guide to Phone-Use. Phone-Use
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗