---

name: guided-qa description: A guided manual testing workflow where the agent pre-verifies features using browser tools, fixes found bugs, and then walks the user through the process.

Guided QA Skill

This skill guides the user through a structured manual testing session. It minimizes user frustration by ensuring the "happy path" works via automated tools before asking the user to perform the steps.

Workflow Overview

1. Scope Definition: Clarify the specific area or feature to be tested if not already stated.
2. Automated Pre-flight: The Agent autonomously walks through the test steps using browser tools (navigate_page, click, fill, etc.) to ensure basic functionality.
3. Pre-flight Fixes: Any bugs encountered during the automated phase are fixed immediately.
4. User Walkthrough: The Agent guides the user step-by-step through the same flow.
5. Interactive Triage: Bugs found by the user are categorized (Blocker vs. Non-Blocker) and handled accordingly.
6. Resolution: Final fixes for non-blocking bugs are applied.

Phase 0: Scope Definition

Goal: Establish exactly what needs to be tested.

1. Check Scope: If the user did not explicitly state what feature or flow to QA in their initial request (e.g., "QA the login flow" vs just "Run QA"), you MUST ask: "What specific area or feature would you like to QA today?"
2. Confirm: Briefly repeat the scope back to the user to ensure alignment before starting Phase 1.

Phase 1: Automated Pre-flight

Goal: Verify the feature works for the Agent before involving the user.

1. Analyze: Review the requested feature set and determine the critical user path.
2. Launch: Ensure the application is running (e.g., just web or just dev-api).
   • Tip: Use run_shell_command to check for running processes/ports.
3. Execute: Use available browser tools to perform the steps.
   • navigate_page: Go to the starting URL (e.g., http://localhost:8081 for Expo Web).
   • take_snapshot: Read the page structure.
   • click / fill: Interact with elements.
   • wait_for: Ensure async operations complete.
4. Fix: If a step fails (e.g., button doesn't work, error message appears):
   • Pause the walkthrough.
   • Investigate the code causing the issue.
   • Fix the bug immediately.
   • Verify the fix by retrying the step.
   • Continue only after the fix works.

Phase 2: User Walkthrough

Goal: Guide the user through the verified path to catch UX issues or edge cases.

1. Instruct: Present one clear action at a time.
   • Bad: "Go to the login page, sign in, and check your profile."
   • Good: "Please navigate to http://localhost:8081." (Wait for confirmation). "Now, enter your email and click Login."
2. Confirm: After each step, ask: "Did that work as expected?" or "What do you see?"
3. Listen: Pay attention to details the user mentions (styling issues, confusion, lag).

Phase 3: Interactive Triage

Goal: Handle issues discovered by the user without derailing the session.

When the user reports a bug, ask: "Is this preventing you from continuing to the next step?"

Scenario A: Blocker (User cannot proceed)

1. Acknowledge: "Understood. This is a blocker."
2. Fix: Switch to coding mode. Locate the issue, fix it, and verify.
3. Resume: Ask the user to retry the step.

Scenario B: Non-Blocker (Visual glitch, minor annoyance)

1. Log: "I've noted that: [Brief description of bug]."
2. Defer: Add it to a QA_FIX_LIST in your memory or a scratchpad.
3. Continue: "Let's proceed to the next step."

Phase 4: Resolution

Goal: Polish the feature by addressing the deferred list.

1. Review: Present the QA_FIX_LIST to the user.
2. Fix: Systematically go through the list and apply fixes.
3. Final Check: Ask the user if they want to verify any of the specific fixes or conclude the session.

Tools Strategy

• Browser Tools: Use navigate_page, take_snapshot, and click heavily during Phase 1.
• Verification: Always use take_snapshot after an action to confirm the UI state changed as expected.
• Debugging: If a selector isn't found, use take_snapshot with verbose=true to debug the accessibility tree.