Agent Browser workflow

Supported command-capable Agents discover and operate Browser Resources through the CLI of the current Farming instance.

Check capability

farming capabilities
farming browser capability
farming browser help workflow

Check current capability before creating a Resource. The existence of a command does not prove that the Browser runtime is available.

Recommended sequence

list → reuse or create → start → navigate → snapshot
     → act from Snapshot references → wait → verify

Observe before acting and verify after acting. Page changes can invalidate old element references and screenshots.

Find Resources

farming browser list

Prefer a matching Browser already owned by the current Agent. Create another Resource only for a separate page or Profile.

If the task needs a page or signed-in session already open in the user's Chrome:

farming browser tabs
farming browser attach <chrome-tab-id>

The Agent selects the matching page by its title and URL. Stopping or deleting the Resource leaves the user's Chrome tab open.

Progressive help

farming browser help navigation
farming browser help interaction
farming browser help inspection
farming browser help debugging
farming browser help files

Read a command's --help, or describe <command> --json, only when exact parameters are needed.

Act and verify

• Wait for an explicit page state after navigation.
• Prefer structured Snapshots for element location.
• Confirm that click and fill targets are unique.
• Observe again after page updates.
• Do not replay a submit, send, or delete operation when a timeout leaves the outcome uncertain.

File boundary

Uploads and downloads remain inside the Browser's Project Workspace. Do not escape through path traversal or silently overwrite existing downloads.

User intervention

For login, CAPTCHA, payment, sensitive data, or irreversible confirmation, stop at a clear point and request user control or explicit authorization. Observe fresh state after control returns.