---
name: "browser-testing-agent-loop"
description: "Exercise a web flow in a real browser and report reproducible failures. Use when the user needs: Get observed results and concise bug reports with relevant evidence."
license: "MIT"
metadata: {"version": "2026-10-17.1", "category": "agents-workflows", "publisher": "AIDiveForge"}
---

# Browser Testing Agent Loop

Get observed results and concise bug reports with relevant evidence.

## Requirements

Requires a browser automation tool and authorization for the tested environment and actions.

## Workflow

1. State the user goal, test account, allowed side effects and starting state. Keep production writes within existing authorization.
2. Run the flow through rendered controls. Check both the visible outcome and relevant requests or persisted state.
3. Exercise desktop and phone layouts when layout is relevant, including keyboard focus and loading/error recovery.
4. Capture only evidence that resolves a concrete issue. Redact tokens and personal data from logs.
5. For each failure give the exact steps, expected/actual behavior, environment and smallest useful evidence.

## Worked example

Read `examples/input.txt` and `examples/expected.txt` before adapting the workflow.

## Acceptance checks

Report the refresh failure separately from the passing search. A mocked click or HTTP 200 alone does not establish rendered success.

## Limits

No browser binary, accounts or hosted testing service is bundled. Results apply only to the environment actually exercised.

Worked example included. Model-specific performance and vendor compatibility have not been benchmarked.