Frontend Debug

Overview

Enable autonomous frontend debugging through empirical browser observation and systematic investigation. This skill implements a structured 6-phase workflow that combines Chrome DevTools MCP for real-world browser testing with automated root cause analysis, fix implementation, and verification. No issue is considered resolved until empirically verified through direct browser interaction.

Core Principle: Evidence-based debugging requires direct browser observation. Never assume a fix works without browser verification.

When to Use This Skill

Invoke this skill for:

• Visual bugs: Elements missing, misaligned, or incorrectly styled
• Interaction failures: Clicks not working, forms not submitting, navigation broken
• Console errors: JavaScript errors, warnings, or unexpected log messages
• Network issues: Failed API requests, incorrect payloads, timeout errors
• Regression detection: Features that stopped working after changes
• GitHub issue debugging: When issue numbers or URLs are referenced

Activation patterns:

# Direct invocation with issue description
@~/.claude/skills/frontend-debug/SKILL.md "Login button not responding"

# With flags for complex issues
@~/.claude/skills/frontend-debug/SKILL.md "Form validation broken" --ultrathink --loop

# From GitHub issue (automatically fetches details)
@~/.claude/skills/frontend-debug/SKILL.md --github-issue 123
@~/.claude/skills/frontend-debug/SKILL.md "#123"  # Shorthand

Workflow Architecture

Execute debugging through 6 systematic phases:

Phase 0: Initialization & Context Gathering Phase 1: Browser Investigation & Observation Phase 2: Root Cause Analysis & Fix Strategy Phase 3: Implementation Phase 4: Empirical Verification (MANDATORY) Phase 5: Reporting & Cleanup

Loop Flag: If

--loop

Phase 0: Initialization & Context Gathering

Objective: Establish debugging session and gather complete issue context.

Step 1: Determine Issue Source

If GitHub issue reference detected (e.g.,

--github-issue 123

"#123"

1. Load

   @contexts/github-integration.md

   for GitHub workflow guidance
2. Fetch issue details via gh CLI:

   gh issue view <number> --json title,body,labels

3. Extract reproduction steps, expected vs actual behavior

If inline description provided:

1. Use description as-is for investigation
2. Ask clarifying questions if ambiguous

If no description:

1. Request user to describe: symptoms, steps to reproduce, expected behavior

Step 2: Detect Project Context

Check for special project configurations:

SoftTrak Detection:

• Check if

  package.json

  contains "softtrak" OR

  .softtrak/

  directory exists
• If detected → Load

  @contexts/softtrak.md

  for credentials and auto-correction details

Git Worktree Detection:

• Run:

  git rev-parse --git-dir | grep "\.git/worktrees"

• If in worktree → Load

  @contexts/worktree.md

  for multi-session management

Step 3: Initialize Session

Create debugging session with tracking:

# Generate session ID
SESSION_ID="{project-name}-{timestamp}"  # or "{worktree-name}-{timestamp}"

# Create checkpoint file for crash recovery
touch .debug-session-${SESSION_ID}.json

# Initialize TodoWrite with 6-phase tracking
# - Phase 0: Initialization ✅
# - Phase 1: Investigation (pending)
# - Phase 2: Root Cause Analysis (pending)
# - Phase 3: Implementation (pending)
# - Phase 4: Verification (pending)
# - Phase 5: Reporting (pending)

Step 4: Check for Incomplete Sessions

Look for existing checkpoint files:

ls -la .debug-session-*.json

If found, offer to resume from last checkpoint with session state.

Checkpoint Criteria: Session initialized, issue understood, TodoWrite created, ready for browser investigation.

Phase 1: Browser Investigation & Observation

Objective: Systematically investigate issue through direct browser observation.

Primary Tool: Chrome DevTools MCP (MANDATORY - always use

--isolated

Investigation Sequence

1. Launch Browser Session

# CRITICAL: Always use --isolated flag to prevent browser conflicts
# Launch new isolated Chrome instance
# Navigate to application URL
# Handle authentication if required (use login helpers if available)
# Navigate to problematic page/feature

2. Capture Initial State

Execute all baseline captures:

required_captures:
  - take_snapshot: Document current DOM/UI state
  - take_screenshot: Visual baseline for comparison
  - list_console_messages: Capture all console output (errors, warnings, info)
  - list_network_requests: Capture all network activity (status codes, payloads)

3. Reproduce Issue

Follow reproduction steps systematically:

• Execute each step from issue description
• Use browser interaction tools:

  click

  ,

  fill

  ,

  hover

  ,

  navigate

• Observe behavior vs expected behavior at each step
• Document exact failure mode and symptoms

4. Deep Investigation

Execute comprehensive checks:

mandatory_checks:
  console_analysis:
    - Identify errors (stack traces, error messages)
    - Note warnings related to issue
    - Check for missing dependencies or failed imports

  network_analysis:
    - Identify failed requests (4xx, 5xx status codes)
    - Inspect request/response payloads
    - Check for timeout errors or slow requests

  dom_analysis:
    - Verify elements exist in DOM via snapshot
    - Check element visibility and styling
    - Validate event listeners are attached

  state_analysis:
    - Inspect localStorage, sessionStorage
    - Check cookies and authentication state
    - Validate application state in console

5. Parallel Analysis (if needed)

For complex issues, invoke additional analysis:

• Sequential MCP: Multi-step reasoning for complex root cause analysis
• Context7 MCP: Framework-specific patterns and debugging approaches
• Codebase Exploration: Read relevant source files for context

Checkpoint Criteria: Issue reproduced in browser, root cause hypothesis formed with confidence score.

Phase 2: Root Cause Analysis & Fix Strategy

Objective: Determine why issue occurs and formulate fix strategy.

Step 1: Invoke Troubleshooting Analysis

Use SuperClaude troubleshoot command:

/sc:troubleshoot [detailed-issue-description] --validate --c7 --seq --persona-analyzer

This provides structured analysis using:

• Sequential MCP for systematic reasoning
• Context7 MCP for framework patterns
• Analyzer persona for root cause investigation

Step 2: Formulate Hypothesis

Structure analysis output:

root_cause_analysis:
  primary_cause: "Why the issue occurs (e.g., event listener not attached)"
  contributing_factors:
    - "Factor 1 (e.g., async timing issue)"
    - "Factor 2 (e.g., missing dependency)"
  confidence_score: 0.75  # 0.0-1.0 scale
  evidence:
    - "Console error showing specific failure"
    - "Network request returning 404"
    - "DOM element missing expected class"

Step 3: Generate Fix Candidates

Develop 1-3 potential fixes ranked by confidence:

fix_candidates:
  option_1:
    approach: "Primary fix strategy"
    confidence: 0.8
    risk: "low"
    rollback: "Easy - single file change"

  option_2:
    approach: "Alternative fix strategy"
    confidence: 0.6
    risk: "medium"
    rollback: "Moderate - multiple files"

Step 4: Request Confirmation (if needed)

If confidence < 70%:

1. Present analysis and proposed fix to user
2. Explain uncertainty and alternatives
3. Request approval before proceeding

Checkpoint Criteria: Root cause identified with acceptable confidence, fix strategy approved.

Step 5: Complexity Escalation Assessment

After root cause analysis, evaluate debugging complexity for potential agent delegation:

Escalation Indicators (if ANY are true, consider agent delegation):

1. Low Confidence: Root cause confidence score <60%
2. Multi-Domain Issue: Multiple interacting systems involved
   • Example: Network (API) + State (Redux) + UI (rendering)
3. Systemic Problem: Issue affects multiple components or features
4. Deep Investigation: Requires analysis across >5 files
5. Specialized Analysis: User explicitly requests domain-specific investigation
6. Multiple Issues: User reports 2+ unrelated problems to debug

Decision Logic:

If escalation indicators detected:

action: delegate_to_agent
reason: "Complex debugging requires specialized sub-agent coordination"
next_step: "Invoke Task tool with frontend-debug-agent"
confidence_requirement: "<60% OR multi_domain:true OR systemic:true"

Agent Invocation Pattern:

Tool: Task
Description: "Complex frontend debugging requiring specialized analysis: [issue-summary]"
Agent Type: root-cause-analyst
Context: {
  browser_evidence: "[screenshots, console logs]",
  reproduction_steps: "[steps from Phase 1]",
  hypothesis: "[from Phase 2 analysis]",
  confidence: "[confidence score]",
  domains_affected: "[network|state|ui|performance]"
}

Agent will spawn specialized debugging sub-agents and coordinate investigation.

If escalation indicators NOT detected:

action: continue_with_skill
reason: "Confidence acceptable, single-domain issue, skill-based debugging appropriate"
next_step: "Proceed to Phase 3: Implementation"

Proceed with skill-based fix implementation:

Phase 3: Implementation

Objective: Apply fix to codebase following best practices.

Step 1: Code Modification

Execute fix using appropriate tools:

file_operations:
  single_file: Use Edit tool
  multiple_files: Use MultiEdit tool for atomic changes

code_standards:
  - Follow existing code patterns and conventions
  - Add inline comments explaining fix
  - Consider edge cases and error handling
  - Maintain consistent formatting

Step 2: Backend Verification (if applicable)

For issues involving backend changes:

# Test API endpoints with curl
curl -X POST http://localhost:3000/api/endpoint \
  -H "Content-Type: application/json" \
  -d '{"test": "data"}'

# Check database state if relevant
# Verify server logs for errors

Step 3: Build & Restart (if required)

Execute build process if needed:

# Rebuild application
npm run build  # or appropriate build command

# Restart dev server
# Clear browser cache if needed
# Verify server is running

Checkpoint Criteria: Fix implemented in code, application ready for empirical testing.

Phase 4: Empirical Verification (MANDATORY)

Objective: Verify fix through direct browser observation.

CRITICAL: This phase is MANDATORY. Never skip browser verification. Use Chrome DevTools MCP exclusively.

Verification Criteria (ALL must pass)

Execute comprehensive verification:

1. UI State Verification ✅

checks:
  - Visual appearance matches expected state
  - Elements present and correctly positioned
  - Styling applied correctly
  - Responsive behavior works
  - Screenshot comparison (before/after)

2. Console Verification ✅

checks:
  - No errors related to the fix
  - No new warnings introduced
  - Expected log messages present
  - Clean console output

3. Network Verification ✅

checks:
  - All requests succeed (200/201/204 status)
  - Payloads correct in structure and content
  - Response times acceptable
  - No timeout errors

4. Interaction Verification ✅

checks:
  - Complete user flow works end-to-end
  - All clickable elements respond
  - Forms submit successfully
  - Navigation functions correctly
  - State persists appropriately

5. Regression Check ✅

checks:
  - Related features still work
  - No new issues introduced
  - Core workflows unaffected
  - Edge cases handled

Verification Process

Execute verification systematically:

step_1_prepare:
  - Hard reload browser (Cmd+Shift+R / Ctrl+Shift+R)
  - Clear cached state if needed
  - Return to starting point of user flow

step_2_execute:
  - Reproduce original issue scenario
  - Observe UI state and behavior at each step
  - Monitor console for errors
  - Monitor network for failures

step_3_validate:
  - Compare actual behavior against expected
  - Take final screenshot for evidence
  - Document verification results
  - Note any unexpected observations

Verification Failure Handling

If ANY verification criterion fails:

With

--loop

Iteration 1 (Skill-based):

• Automatically restart from Phase 1
• Re-investigate with new observations in skill context

If Iteration 1 fails → Escalation Decision:

iteration_1_failed: true
action: "Evaluate complexity for agent delegation"
threshold: "If confidence <50% OR new domains discovered → delegate"

Iterations 2-3 (Agent-based if escalated):

• Invoke frontend-debug-agent for deeper investigation
• Agent spawns specialized sub-agents as needed
• Agent coordinates up to 2 additional iterations
• Maximum total: 1 skill iteration + 2 agent iterations = 3 total

Rationale: Keep simple retries in main context, escalate complex cases to agents.

Without

--loop

1. Report verification failure with details
2. Explain which criteria failed and why
3. Request user guidance on next steps

Checkpoint Criteria: All 5 verification criteria passed, fix confirmed working.

Phase 5: Reporting & Cleanup

Objective: Document findings and close debugging session.

Step 1: Generate Investigation Report

Create comprehensive report:

# Frontend Debug Report: {Issue Title}

## Summary
[One-line description of issue and fix]

## Root Cause
[Detailed explanation of why issue occurred]

## Fix Applied
[What was changed and why it resolves the issue]

## Verification Results
- ✅ UI State: [Details]
- ✅ Console: [Details]
- ✅ Network: [Details]
- ✅ Interactions: [Details]
- ✅ Regression: [Details]

## Evidence
- Before Screenshot: [path]
- After Screenshot: [path]
- Console Logs: [relevant logs]
- Network Activity: [relevant requests]

## Recommended Tests
[Suggested regression tests to prevent future issues]

## Session Details
- Session ID: {session-id}
- Duration: {elapsed-time}
- Iterations: {loop-count}

Step 2: GitHub Integration (if applicable)

If debugging from GitHub issue:

# Post resolution comment with report
gh issue comment <number> --body "[report-content]"

# Add resolved label to indicate fix has been applied and verified
gh issue edit <number> --add-label "resolved-by-claude"

# Remove automated-debug label if present
gh issue edit <number> --remove-label "automated-debug"

# DO NOT automatically close the issue
# Let human developers review the fix and close manually
# This ensures proper validation and prevents premature closure

Step 3: Session Cleanup

Clean up debugging artifacts:

# Mark all TodoWrite tasks as completed
# Delete checkpoint file: rm .debug-session-${SESSION_ID}.json
# Close browser session via Chrome DevTools MCP
# Save final report to appropriate location

Step 4: Knowledge Base Update (optional but recommended)

Store learning for future debugging:

{
  "issue_pattern": "description of issue type",
  "root_cause_category": "async timing | network | state management | etc",
  "solution_pattern": "approach that resolved issue",
  "detection_signals": ["symptom 1", "symptom 2"],
  "fix_template": "reusable solution approach"
}

Checkpoint Criteria: Report generated, GitHub issue updated (if applicable), session cleaned up.

Error Handling

Browser Session Failures

chrome_devtools_unavailable:
  symptom: "MCP connection fails"
  action: "Check Chrome DevTools MCP configuration in .mcp.json"
  fallback: "Request user to verify MCP setup and Chrome installation"

browser_already_running:
  symptom: "Port conflict or existing instance"
  cause: "--isolated flag not used"
  solution: "Always use --isolated=true for new debugging sessions"

navigation_timeout:
  symptom: "Page load timeout"
  action: "Increase timeout or check network connectivity"
  fallback: "Request user to verify application is running and accessible"

authentication_required:
  symptom: "Login page or auth wall"
  action: "Use project-specific login helper if available"
  fallback: "Request user credentials or authentication approach"

Verification Failures

fix_does_not_work:
  with_loop_flag:
    action: "Automatically restart Phase 1 (max 3 iterations)"
    logging: "Document why previous fix failed"

  without_loop_flag:
    action: "Report failure with detailed explanation"
    next_step: "Request user guidance on alternative approaches"

new_issue_introduced:
  action: "Rollback changes to previous working state"
  logging: "Document regression details"
  next_step: "Re-analyze with different fix strategy"

partial_success:
  action: "Document which criteria passed vs failed"
  decision: "Discuss with user whether partial fix is acceptable"

GitHub Integration Failures

gh_cli_not_available:
  symptom: "gh command not found"
  action: "Warn user to install: brew install gh"
  fallback: "Continue with manual issue description"

issue_not_found:
  symptom: "Issue number doesn't exist"
  action: "Verify issue number and repository"
  fallback: "Request user to provide issue details manually"

authentication_failed:
  symptom: "gh auth required"
  action: "Guide user through: gh auth login"
  fallback: "Continue without GitHub integration"

Tool Reference Failures

If encountering errors with Chrome DevTools MCP tools:

1. Load

   @references/mcp-tools.md

   for detailed tool documentation
2. Review proper tool usage and parameters
3. Check for MCP configuration issues

Conditional Context Loading

Design Principle: Load additional context only when detected or needed to optimize token usage.

When to Load Additional Context

GitHub Integration → Load

@contexts/github-integration.md

• Trigger:

  --github-issue

  flag OR GitHub issue reference detected (#123, URLs)
• Provides: Issue lifecycle management, gh CLI commands, automated tracking
• Token Cost: +1,000 tokens

SoftTrak Project → Load

@contexts/softtrak.md

• Trigger:

  package.json

  contains "softtrak" OR

  .softtrak/

  directory exists
• Provides: Test credentials, auto-correction, project-specific configurations
• Token Cost: +600 tokens

Git Worktree → Load

@contexts/worktree.md

• Trigger:

  git rev-parse --git-dir | grep "\.git/worktrees"

  succeeds
• Provides: Multi-session management, browser isolation strategies
• Token Cost: +400 tokens

Tool Reference → Load

@references/mcp-tools.md

• Trigger: Agent encounters tool errors OR needs specific tool guidance
• Provides: Detailed Chrome DevTools MCP tool reference with examples
• Token Cost: +2,000 tokens

Usage Examples → Load

@references/examples.md

• Trigger: User requests examples OR agent needs workflow clarification
• Provides: Example scenarios, expected workflows, common patterns
• Token Cost: +1,500 tokens

Resources

This skill includes organized resources for effective debugging:

scripts/

cleanup-sessions.sh

• Remove stale debug session checkpoint files
• Execute periodically to prevent workspace clutter
• Usage:

  bash scripts/cleanup-sessions.sh

references/

mcp-tools.md

• Load when encountering tool errors
• Provides comprehensive tool usage examples
• Includes parameter documentation

examples.md

• Load when needing workflow clarification
• Shows realistic debugging sessions
• Demonstrates best practices

contexts/

github-integration.md

• Load when

  --github-issue

  flag used or issue reference detected
• Provides gh CLI command reference
• Documents automated tracking workflow

softtrak.md

• Load when SoftTrak project detected
• Provides test credentials and auto-correction details
• Documents project-specific patterns

worktree.md

• Load when in git worktree environment
• Provides browser isolation strategies
• Documents concurrent debugging workflows

templates/

investigation-report-template.md

verification-checklist.md

session-state-schema.json

mcp-config-example.json

Knowledge Base

knowledge-base.json

framework-quirks.json

project-contexts/softtrak.json

Best Practices

1. Always Use Chrome DevTools MCP: Empirical verification requires real browser observation
2. Always Use --isolated Flag: Prevents browser conflicts in concurrent debugging sessions
3. Load Context Conditionally: Only load additional modules when detection patterns match
4. Document Thoroughly: Maintain detailed checkpoint files for crash recovery
5. Verify Empirically: Never assume a fix works without browser verification in Phase 4
6. Loop When Uncertain: Use

   --loop

   flag for complex issues requiring iteration
7. GitHub Integration: Use issue tracking for accountability and knowledge retention
8. Session Management: Create checkpoint files for recovery from crashes or interruptions

Performance & Token Optimization

Modular Loading Strategy: Conditional loading minimizes token usage.

Token Budget:

• Core Workflow (SKILL.md): ~4,500 tokens
• GitHub Integration: +1,000 tokens (conditional)
• SoftTrak Context: +600 tokens (conditional)
• Worktree Support: +400 tokens (conditional)
• Tool Reference: +2,000 tokens (conditional)
• Usage Examples: +1,500 tokens (conditional)

Total Range: 4,500 tokens (minimal) to 10,000 tokens (all modules loaded)

Optimization: Load only what's needed based on detection patterns, achieving 30-70% token savings vs loading everything upfront.