Skip to content

Latest commit

 

History

History
48 lines (39 loc) · 1.79 KB

File metadata and controls

48 lines (39 loc) · 1.79 KB

AWT (AI Watch Tester) — AI Agent Rules

This file is for AI coding assistants (Gemini, Copilot, Codex, Cursor, Claude Code, etc.). Follow these rules when the user asks you to use AWT for testing.

⛔ MANDATORY WORKFLOW — 4 Steps, No Exceptions

When the user asks you to test a web application with AWT, you MUST follow these 4 steps in order. Do NOT skip, combine, or automate any step.

Step 1: SCAN

aat scan --url <URL>

Read .aat/scan_result.json and present a summary to the user:

  • How many elements were found (inputs, buttons, links)
  • Ask: "Should I create a test scenario based on these elements?"
  • WAIT for user response.

Step 2: GENERATE + PRESENT

Write a YAML scenario using data from scan_result.json. Show the full scenario to the user in a readable format. Ask: "Should I run this test? Or do you want me to change anything?"

  • WAIT for user approval.
  • If user requests changes → modify and show again.

Step 3: EXECUTE (only after user says "go ahead")

aat run --skill-mode --fast <scenario_file>

If a step fails → STOP immediately. Report:

  • Which step failed and why
  • What the possible cause is
  • Ask: "Should I fix the scenario, or fix the source code?"
  • WAIT for user instruction. Do NOT auto-fix.

Step 4: REPORT

When all steps pass, summarize the results to the user.

⛔ BANNED — Never Do These

Banned Why
aat devqa Runs entire pipeline without user checkpoints
-y / --auto-approve Bypasses user approval gate
Auto-fixing code without asking User must approve all changes
Auto-retrying failed tests User must decide what to fix first
headless: true User must see the browser
Guessing element names Always scan first, use real data