quick-debug

---

name: quick-debug description: On test failure, jump directly to the problem page via token login and debug in-place. Trigger: debug, troubleshoot, locator failure, timeout, element not found, assertion failure.

Quick Debug — Rapid Triage and Fix

Core Idea

Skip the full flow from login to the problem node; go directly to the target page via token login-free access and debug in-place.

Leverage the project's existing token URL login-free mechanism, combined with browser tools, to perform DOM exploration, locator verification, and wait strategy adjustment directly on the problem page. Fix and verify in-place without re-running the full test flow.

---

Main Flow (5 Steps)

Step 1: Parse failure information
    ↓
Step 2: Construct login-free redirect URL
    ↓
Step 3: Redirect to target page without login
    ↓
Step 4: Diagnose the problem (decision tree)
    ↓
Step 5: Fix + in-place verification

Step 1: Parse Failure Information

Extract from the stack trace / user description: failed test name, error type (TimeoutError / AssertionError), failed locator, URL at time of failure. If information is insufficient, read the test code to infer.

Step 2: Construct Login-Free Redirect URL

Read the token for the corresponding environment from the config file.

⚠️ Must confirm --env with the user before login-free access (see agent-behavior P0.2).

Two cases:

Case	Strategy
Target page has a clear URL path	Login-free first (/entry?token=<JWT>), then navigate to target path
Target page requires interaction to reach (dialog, new tab, etc.)	Login-free to the nearest directly-accessible parent page, then perform minimum navigation to reach the target

Step 3: Redirect to Target Page Without Login

browser_navigate → <base_url>/entry?token=<JWT> → navigate to target path → verify is_page_loaded(). If login-free fails (redirected back to login page) → prompt user to refresh token.

Step 4: Diagnose the Problem

See the "Diagnosis Decision Tree" section below.

Step 5: Fix + In-Place Verification

Call the corresponding skill based on diagnosis result:

Diagnosis Result	Call Skill
Locator issue	locator-replacer
Wait strategy issue	wait-strategy
Page load assertion issue	page-load-assertion
Major page redesign	Report to user, suggest gen-page-test to regenerate
Assertion expected value changed	Report the difference, let user decide whether to update expected values

In-place verification: After fixing, directly verify on the current browser page (re-execute the failed operation with Playwright MCP). Once confirmed, suggest that the user re-run the full tests.

---

Diagnosis Decision Tree (Step 4 Expanded)

First use agent-browser to quickly scan the current page DOM (fall back to Playwright MCP if not installed), then follow the corresponding branch based on error type:

Error type classification
    ├── A. Locator issue (element not found / matches multiple)
    ├── B. Timeout issue (element exists but loads slowly / network latency)
    ├── C. Assertion issue (element exists but content/state doesn't match expectation)
    └── D. Page structure change (overall page redesign / route change)

Branch A: Locator Issue

Criteria: locator.count() == 0 or strict mode violation (multiple matches)

Investigation steps:

1. Use agent-browser to capture a DOM snapshot of the problem area
2. Compare PageObject locator with actual DOM, determine cause: class name change / text change / dynamic rendering not complete

Fix path → call locator-replacer skill

Branch B: Timeout Issue

Criteria: TimeoutError and element can appear after a longer wait, or networkidle not reached

Investigation steps:

1. Check if current page URL is correct (rule out navigation deviation); check if target element exists but is not visible (hidden / covered)
2. Determine cause: implicit wait insufficient / SPA render delay / slow network interface

Fix path → call wait-strategy skill or adjust timeout configuration

Branch C: Assertion Issue

Criteria: AssertionError, element exists and is visible, but value/state doesn't match expectation

Investigation steps:

1. Get actual element text/attribute value, compare with expected value in test code
2. Determine cause: business data change / copy update / environment difference

Fix path → report the difference and suggest user decision (usually a business change rather than a code bug)

Branch D: Page Structure Change

Criteria: is_page_loaded() fails or current URL doesn't match expectation

Investigation steps:

1. Use agent-browser to capture a full-page snapshot, batch-compare with PageObject locators
2. Determine cause: route change / page redesign / permission change causing redirect

Fix path → for small changes call locator-replacer; for large redesigns report to user, suggest gen-page-test

Fallback Mechanism

If classification-based investigation yields no result, fall back to the previous level and reclassify. Maximum one fallback; if still unable to locate, compile collected information and report to user.

---

Browser Tool Selection

DOM exploration/locator collection → agent-browser (preferred); precise verification/assertions → Playwright MCP; if agent-browser not installed → use Playwright MCP throughout (P0.4 fallback).

---

Checklist

Before each quick-debug session ends, confirm:

• Root cause identified and classified (which branch A/B/C/D)
• Fix action executed (or reported to user for manual decision)
• In-place verification passed (fixed locator/wait is effective on current page)
• Suggested user re-run full tests to confirm regression
• For save operation verification failures, refer to save-verify-strategy (Toast timing / redirect detection / data comparison)