GTM Performance Audit

Analyze Google Tag Manager implementations for performance impact using Chrome DevTools MCP.

Philosophy

Tag managers are powerful but can silently degrade performance. Every tag has a cost.

Core Beliefs

Measure Tag Impact Individually: Each tag's cost must be measured, not assumed
Container Size Is a Leading Indicator: Large containers correlate with poor performance
Triggers Are the Multiplier: Inefficient triggers amplify tag cost exponentially
Server-Side Is the Future: Move data collection off the main thread when possible

Why GTM Audits Matter

Invisible Performance Tax: Tags add up silently; no single tag seems costly
Core Web Vitals Impact: GTM scripts affect LCP, INP, and CLS directly
Third-Party Risk: Tags load external resources outside your control
Consent Compliance: Tags firing before consent degrade trust and may violate regulations

When This Skill Activates

Activate this skill when the user:

Mentions "GTM", "Google Tag Manager", or "tag manager"
Says "too many tags" or "tracking is slow"
Asks "are our marketing tags slowing the site?"
Mentions "container size" or "tag execution"
References "dataLayer", "custom HTML tags", or "trigger efficiency"
Asks about "tag performance" or "third-party script impact"
Shows GTM-related code and asks about performance

Decision Framework

Before auditing, determine the audit mode:

What Data Is Available?

Live page with GTM → Full audit via Chrome DevTools MCP (best)
Container JSON export + live page → Container analysis + live profiling (comprehensive)
Container ID only → Infer from network requests on live page (limited)
Code review only → Analyze GTM integration code in codebase (minimal)

What's the Goal?

Quick check → Container size, tag count, obvious issues
Standard audit → Full analysis with CWV impact mapping
Comprehensive → Deep profiling with remediation plan and implementation steps

Decision Tree

User asks about GTM performance
    |
    v
URL provided? → NO → Ask for target URL (STOP until answered)
    |
   YES
    |
    v
Proceed autonomously (no further permission prompts needed)
    |
    v
Phase 1: Baseline Performance (before GTM analysis)
    |
    v
Phase 2: GTM Container Analysis
    |
    v
Phase 3: Issue Detection (14 checks)
    |
    v
Phase 4: Report Generation

Setup

Before starting the audit, gather from the user:

Target URL (required) - The page to audit
GTM Container ID (required) - Format: GTM-XXXXXXX
Container JSON export (optional) - Exported from GTM admin
Critical tags list (optional) - Tags that must not be removed

Phase 1: Baseline Performance

Measure page performance using the single-pass audit collector script — one evaluate_script call replaces all individual measurement scripts.

Step 1: Read the Collector Script

Read: agents/gtm-specialist/scripts/gtm-audit-collector.js

Step 2: Navigate and Run the Collector

mcp__chrome-devtools__navigate_page(url="<target-url>")

Wait for the page to fully load, then execute:

mcp__chrome-devtools__evaluate_script(function="<full content of gtm-audit-collector.js>")

This single call returns a structured JSON object containing:

Navigation timing (TTFB, DCL, load complete)
Paint metrics (FCP, first paint)
Core Web Vitals (LCP with element, CLS, INP estimate, with ratings)
Long Tasks / Total Blocking Time
All <head> scripts with blocking/async/defer status
All GTM and third-party tracking resources (40+ vendors)
DataLayer contents and unique events
Consent platform detection (Google Consent Mode v2, TrustArc, OneTrust, etc.)
Window-global vendor fingerprinting (60+ vendors)
GA4 Measurement IDs, Conversion IDs, legacy UA IDs

Step 3: Supplemental Data (if needed)

Run these only if the collector output is insufficient for a specific check:

Console errors (check for GTM JS errors):

mcp__chrome-devtools__list_console_messages()

Raw network waterfall (for precise HAR timing):

mcp__chrome-devtools__list_network_requests()

Deep CPU profiling (comprehensive depth mode only):

mcp__chrome-devtools__performance_start_trace(reload=true, autoStop=true)
mcp__chrome-devtools__performance_stop_trace()

Phase 2: GTM Container Analysis

If Container JSON Available

Parse the exported JSON to extract:

Container size (bytes)
Tag count by type (Custom HTML, Google Analytics, Floodlight, etc.)
Trigger count and types (Page View, Click, Custom Event, etc.)
Variable count and types (DataLayer, DOM Element, JavaScript, etc.)
Tag firing sequences and dependencies
Custom HTML content (inline scripts)

If Only Container ID Available

Infer container details from network data:

Count tags from network requests
Identify third-party domains loaded
Measure script sizes and timing
Check for known tag patterns (GA4, Meta Pixel, etc.)

DataLayer Analysis

// Use mcp__chrome-devtools__evaluate_script
() => {
  if (typeof dataLayer === 'undefined') return { error: 'No dataLayer found' };
  return {
    eventCount: dataLayer.length,
    events: dataLayer.map(item => ({
      event: item.event || 'push',
      keys: Object.keys(item)
    })),
    totalSize: JSON.stringify(dataLayer).length
  };
}

Phase 3: Issue Detection

Run all 14 check categories:

1. Synchronous Script Loading

What: GTM container loaded with sync instead of async Impact: Blocks HTML parsing, delays LCP Check: Look for <script src="...gtm.js"> without async or defer Fix: Ensure GTM snippet uses async loading pattern

2. Blocking Tags

What: Tags that block the main thread for >50ms Impact: Increases INP, degrades interactivity Check: Profile main thread during tag execution Fix: Defer non-critical tags, use tag sequencing

3. Large Payloads

What: Tags or container >100KB compressed Impact: Increases load time, wastes bandwidth Check: Measure transfer sizes of GTM resources Fix: Remove unused tags, optimize custom HTML, consider server-side

4. Main Thread Blocking

What: JavaScript execution blocks main thread Impact: Poor INP, delayed interaction response Check: Use performance trace to measure long tasks Fix: Break up long-running tag scripts, use Web Workers where possible

5. Missing Conditional Firing

What: Tags fire on all pages when only needed on specific pages Impact: Unnecessary processing and network requests Check: Analyze trigger conditions vs. page types Fix: Add page path or event conditions to triggers

6. Trigger Optimization

What: Triggers using expensive DOM selectors or all-pages rules Impact: Excessive event listener overhead Check: Review trigger configurations for broad matching Fix: Use specific CSS selectors, limit to relevant pages

7. Duplicate Tags

What: Same tracking pixel or analytics tag firing multiple times Impact: Inflated metrics, wasted resources, double-counting Check: Compare tag endpoints and parameters Fix: Remove duplicates, consolidate similar tags

8. Orphaned Tags

What: Tags with no active triggers (paused or misconfigured) Impact: Container bloat, maintenance burden Check: Cross-reference tags with their trigger assignments Fix: Remove or archive orphaned tags

9. Expensive Variables

What: Variables using DOM lookups or complex JavaScript on every evaluation Impact: Repeated expensive computations Check: Review variable types and computation cost Fix: Cache values in dataLayer, use simpler variable types

10. Missing Async Attribute

What: Third-party scripts injected by tags without async/defer Impact: Render-blocking, delays page load Check: Inspect Custom HTML tags for script injection patterns Fix: Add async attribute to injected scripts

11. Custom HTML Best Practices

What: Custom HTML tags with inline scripts that could use built-in tag types Impact: Security risk, harder to maintain, potentially blocking Check: Analyze Custom HTML content for convertible patterns Fix: Convert to built-in tag templates where possible

12. Server-Side Candidates

What: Tags that could run server-side instead of client-side Impact: Reduces client-side JavaScript, improves performance Check: Identify tags sending data to first-party or server endpoints Fix: Migrate eligible tags to server-side GTM container

13. Consent Mode Gaps

What: Tags firing before consent is granted Impact: Privacy violations, regulatory risk Check: Verify consent mode integration and tag firing sequence Fix: Implement Google Consent Mode, add consent triggers

14. Tag Firing Order

What: Critical tags delayed by non-essential tags in the sequence Impact: Important data collection delayed Check: Review tag priority settings and sequencing Fix: Set proper priority values, use tag sequencing groups

Phase 4: Report Generation

Executive Summary

## GTM Performance Audit Summary

**Container**: GTM-XXXXXXX
**URL Tested**: https://example.com
**Date**: YYYY-MM-DD

### Key Metrics
- Container Size: XXX KB (compressed)
- Total Tags: NN (XX active, YY paused)
- GTM Load Time: XXXms
- Main Thread Blocking: XXXms
- Estimated CWV Impact: [Low/Medium/High]

### Quick Wins
1. [Quick win with estimated impact]
2. [Quick win with estimated impact]
3. [Quick win with estimated impact]

Detailed Findings

Each finding should include:

Description: What the issue is
Measurement: Quantified impact (ms, KB, count)
CWV Impact: Which Core Web Vital is affected and by how much
GTM Fix Steps: Specific steps to fix in GTM admin
Risk Assessment: Low/Medium/High risk of the fix
Priority: Critical/High/Medium/Low

Tag Inventory Table

| Tag Name | Type | Trigger | Size | Exec Time | Status |
|----------|------|---------|------|-----------|--------|
| GA4 | Google Analytics | All Pages | 45KB | 120ms | Active |
| Meta Pixel | Custom HTML | All Pages | 32KB | 85ms | Active |
| Hotjar | Custom HTML | Page View | 28KB | 200ms | Active |

Network Waterfall

Show timing of GTM-related requests:

0ms     100ms    200ms    300ms    400ms    500ms
|--------|--------|--------|--------|--------|
[==GTM Container (45KB)==]
         [====GA4 (32KB)====]
         [==Meta Pixel==]
                  [====Hotjar (28KB)====]
                           [=LinkedIn=]

Implementation Checklist

## Remediation Checklist

### Critical (Do This Week)
- [ ] Remove duplicate GA4 tag (saves ~120ms)
- [ ] Add async to Custom HTML script injections

### High Priority (Do This Sprint)
- [ ] Add page path conditions to 5 all-pages tags
- [ ] Convert 3 Custom HTML tags to built-in templates
- [ ] Remove 4 orphaned tags (saves ~15KB)

### Medium Priority (Plan for Next Sprint)
- [ ] Evaluate server-side GTM for conversion tags
- [ ] Implement consent mode v2
- [ ] Optimize dataLayer push frequency

### Low Priority (Backlog)
- [ ] Consolidate similar event triggers
- [ ] Review variable caching strategy

Required MCP Integration

This skill requires Chrome DevTools MCP. Tools used:

mcp__chrome-devtools__evaluate_script - Execute JavaScript for timing, dataLayer, resource analysis
mcp__chrome-devtools__list_network_requests - Capture GTM network waterfall
mcp__chrome-devtools__list_console_messages - Check for GTM errors
mcp__chrome-devtools__navigate_page - Navigate to target URL
mcp__chrome-devtools__take_snapshot - Get page structure for tag detection
mcp__chrome-devtools__performance_start_trace - Start performance profiling
mcp__chrome-devtools__performance_stop_trace - Stop profiling and get results
mcp__chrome-devtools__new_page - Open fresh tab for clean measurement

CMS-Specific GTM Patterns

Drupal

Common GTM Integration:

google_tag module (most common)
Custom html.html.twig injection
Config at /admin/config/services/google-tag

Check for:

Module configuration: drush config:get google_tag.settings
Container ID in config vs. hardcoded in templates
Multiple containers (module + theme + custom module)
DataLayer integration via hook_google_tag_snippets_alter()

WordPress

Common GTM Integration:

GTM4WP plugin (most common)
Google Site Kit plugin
Manual header.php injection
Theme functions.php via wp_head action

Check for:

Plugin settings: GTM4WP → Settings → Container ID
Multiple injection points (plugin + theme + another plugin)
DataLayer population via WooCommerce integration
Custom event tracking in theme JavaScript

Guardrails

Read-only: Never modify GTM containers, tags, triggers, or variables
No credentials: Never ask for or enter GTM admin credentials
Conservative estimates: When unsure of impact, estimate conservatively
Preserve functionality: Recommendations must not break tracking
Privacy-aware: Flag consent issues but don't disable tags

Integration with /audit-gtm Command

This Skill: Focused GTM analysis during conversation
- "Are our marketing tags slow?"
- "How big is our GTM container?"
- Single-tag or quick analysis
/audit-gtm Command: Comprehensive GTM performance audit
- Full container analysis
- All 14 check categories
- Detailed report with remediation plan

Quick Tips

Container under 50KB compressed is healthy; over 100KB needs attention
More than 30 active tags usually indicates cleanup opportunity
Custom HTML tags are the biggest performance risk - audit these first
All Pages triggers are the most common source of waste
dataLayer pushes before GTM loads are queued and processed in order

GTM Performance Audit

Analyze Google Tag Manager implementations for performance impact using Chrome DevTools MCP.

Philosophy

Tag managers are powerful but can silently degrade performance. Every tag has a cost.

Core Beliefs

Measure Tag Impact Individually: Each tag's cost must be measured, not assumed
Container Size Is a Leading Indicator: Large containers correlate with poor performance
Triggers Are the Multiplier: Inefficient triggers amplify tag cost exponentially
Server-Side Is the Future: Move data collection off the main thread when possible

Why GTM Audits Matter

Invisible Performance Tax: Tags add up silently; no single tag seems costly
Core Web Vitals Impact: GTM scripts affect LCP, INP, and CLS directly
Third-Party Risk: Tags load external resources outside your control
Consent Compliance: Tags firing before consent degrade trust and may violate regulations

When This Skill Activates

Activate this skill when the user:

Mentions "GTM", "Google Tag Manager", or "tag manager"
Says "too many tags" or "tracking is slow"
Asks "are our marketing tags slowing the site?"
Mentions "container size" or "tag execution"
References "dataLayer", "custom HTML tags", or "trigger efficiency"
Asks about "tag performance" or "third-party script impact"
Shows GTM-related code and asks about performance

Decision Framework

Before auditing, determine the audit mode:

What Data Is Available?

Live page with GTM → Full audit via Chrome DevTools MCP (best)
Container JSON export + live page → Container analysis + live profiling (comprehensive)
Container ID only → Infer from network requests on live page (limited)
Code review only → Analyze GTM integration code in codebase (minimal)

What's the Goal?

Quick check → Container size, tag count, obvious issues
Standard audit → Full analysis with CWV impact mapping
Comprehensive → Deep profiling with remediation plan and implementation steps

Decision Tree

User asks about GTM performance
    |
    v
URL provided? → NO → Ask for target URL (STOP until answered)
    |
   YES
    |
    v
Proceed autonomously (no further permission prompts needed)
    |
    v
Phase 1: Baseline Performance (before GTM analysis)
    |
    v
Phase 2: GTM Container Analysis
    |
    v
Phase 3: Issue Detection (14 checks)
    |
    v
Phase 4: Report Generation

Setup

Before starting the audit, gather from the user:

Target URL (required) - The page to audit
GTM Container ID (required) - Format: GTM-XXXXXXX
Container JSON export (optional) - Exported from GTM admin
Critical tags list (optional) - Tags that must not be removed

Phase 1: Baseline Performance

Measure page performance using the single-pass audit collector script — one evaluate_script call replaces all individual measurement scripts.

Step 1: Read the Collector Script

Read: agents/gtm-specialist/scripts/gtm-audit-collector.js

Step 2: Navigate and Run the Collector

mcp__chrome-devtools__navigate_page(url="<target-url>")

Wait for the page to fully load, then execute:

mcp__chrome-devtools__evaluate_script(function="<full content of gtm-audit-collector.js>")

This single call returns a structured JSON object containing:

Navigation timing (TTFB, DCL, load complete)
Paint metrics (FCP, first paint)
Core Web Vitals (LCP with element, CLS, INP estimate, with ratings)
Long Tasks / Total Blocking Time
All <head> scripts with blocking/async/defer status
All GTM and third-party tracking resources (40+ vendors)
DataLayer contents and unique events
Consent platform detection (Google Consent Mode v2, TrustArc, OneTrust, etc.)
Window-global vendor fingerprinting (60+ vendors)
GA4 Measurement IDs, Conversion IDs, legacy UA IDs

Step 3: Supplemental Data (if needed)

Run these only if the collector output is insufficient for a specific check:

Console errors (check for GTM JS errors):

mcp__chrome-devtools__list_console_messages()

Raw network waterfall (for precise HAR timing):

mcp__chrome-devtools__list_network_requests()

Deep CPU profiling (comprehensive depth mode only):

mcp__chrome-devtools__performance_start_trace(reload=true, autoStop=true)
mcp__chrome-devtools__performance_stop_trace()

Phase 2: GTM Container Analysis

If Container JSON Available

Parse the exported JSON to extract:

Container size (bytes)
Tag count by type (Custom HTML, Google Analytics, Floodlight, etc.)
Trigger count and types (Page View, Click, Custom Event, etc.)
Variable count and types (DataLayer, DOM Element, JavaScript, etc.)
Tag firing sequences and dependencies
Custom HTML content (inline scripts)

If Only Container ID Available

Infer container details from network data:

Count tags from network requests
Identify third-party domains loaded
Measure script sizes and timing
Check for known tag patterns (GA4, Meta Pixel, etc.)

DataLayer Analysis

// Use mcp__chrome-devtools__evaluate_script
() => {
  if (typeof dataLayer === 'undefined') return { error: 'No dataLayer found' };
  return {
    eventCount: dataLayer.length,
    events: dataLayer.map(item => ({
      event: item.event || 'push',
      keys: Object.keys(item)
    })),
    totalSize: JSON.stringify(dataLayer).length
  };
}

Phase 3: Issue Detection

Run all 14 check categories:

1. Synchronous Script Loading

2. Blocking Tags

3. Large Payloads

4. Main Thread Blocking

5. Missing Conditional Firing

6. Trigger Optimization

7. Duplicate Tags

8. Orphaned Tags

9. Expensive Variables

10. Missing Async Attribute

11. Custom HTML Best Practices

12. Server-Side Candidates

13. Consent Mode Gaps

14. Tag Firing Order

Phase 4: Report Generation

Executive Summary

## GTM Performance Audit Summary

**Container**: GTM-XXXXXXX
**URL Tested**: https://example.com
**Date**: YYYY-MM-DD

### Key Metrics
- Container Size: XXX KB (compressed)
- Total Tags: NN (XX active, YY paused)
- GTM Load Time: XXXms
- Main Thread Blocking: XXXms
- Estimated CWV Impact: [Low/Medium/High]

### Quick Wins
1. [Quick win with estimated impact]
2. [Quick win with estimated impact]
3. [Quick win with estimated impact]

Detailed Findings

Each finding should include:

Description: What the issue is
Measurement: Quantified impact (ms, KB, count)
CWV Impact: Which Core Web Vital is affected and by how much
GTM Fix Steps: Specific steps to fix in GTM admin
Risk Assessment: Low/Medium/High risk of the fix
Priority: Critical/High/Medium/Low

Tag Inventory Table

| Tag Name | Type | Trigger | Size | Exec Time | Status |
|----------|------|---------|------|-----------|--------|
| GA4 | Google Analytics | All Pages | 45KB | 120ms | Active |
| Meta Pixel | Custom HTML | All Pages | 32KB | 85ms | Active |
| Hotjar | Custom HTML | Page View | 28KB | 200ms | Active |

Network Waterfall

Show timing of GTM-related requests:

0ms     100ms    200ms    300ms    400ms    500ms
|--------|--------|--------|--------|--------|
[==GTM Container (45KB)==]
         [====GA4 (32KB)====]
         [==Meta Pixel==]
                  [====Hotjar (28KB)====]
                           [=LinkedIn=]

Implementation Checklist

## Remediation Checklist

### Critical (Do This Week)
- [ ] Remove duplicate GA4 tag (saves ~120ms)
- [ ] Add async to Custom HTML script injections

### High Priority (Do This Sprint)
- [ ] Add page path conditions to 5 all-pages tags
- [ ] Convert 3 Custom HTML tags to built-in templates
- [ ] Remove 4 orphaned tags (saves ~15KB)

### Medium Priority (Plan for Next Sprint)
- [ ] Evaluate server-side GTM for conversion tags
- [ ] Implement consent mode v2
- [ ] Optimize dataLayer push frequency

### Low Priority (Backlog)
- [ ] Consolidate similar event triggers
- [ ] Review variable caching strategy

Required MCP Integration

This skill requires Chrome DevTools MCP. Tools used:

mcp__chrome-devtools__evaluate_script - Execute JavaScript for timing, dataLayer, resource analysis
mcp__chrome-devtools__list_network_requests - Capture GTM network waterfall
mcp__chrome-devtools__list_console_messages - Check for GTM errors
mcp__chrome-devtools__navigate_page - Navigate to target URL
mcp__chrome-devtools__take_snapshot - Get page structure for tag detection
mcp__chrome-devtools__performance_start_trace - Start performance profiling
mcp__chrome-devtools__performance_stop_trace - Stop profiling and get results
mcp__chrome-devtools__new_page - Open fresh tab for clean measurement

CMS-Specific GTM Patterns

Drupal

Common GTM Integration:

google_tag module (most common)
Custom html.html.twig injection
Config at /admin/config/services/google-tag

Check for:

Module configuration: drush config:get google_tag.settings
Container ID in config vs. hardcoded in templates
Multiple containers (module + theme + custom module)
DataLayer integration via hook_google_tag_snippets_alter()

WordPress

Common GTM Integration:

GTM4WP plugin (most common)
Google Site Kit plugin
Manual header.php injection
Theme functions.php via wp_head action

Check for:

Plugin settings: GTM4WP → Settings → Container ID
Multiple injection points (plugin + theme + another plugin)
DataLayer population via WooCommerce integration
Custom event tracking in theme JavaScript

Guardrails

Read-only: Never modify GTM containers, tags, triggers, or variables
No credentials: Never ask for or enter GTM admin credentials
Conservative estimates: When unsure of impact, estimate conservatively
Preserve functionality: Recommendations must not break tracking
Privacy-aware: Flag consent issues but don't disable tags

Integration with /audit-gtm Command

This Skill: Focused GTM analysis during conversation
- "Are our marketing tags slow?"
- "How big is our GTM container?"
- Single-tag or quick analysis
/audit-gtm Command: Comprehensive GTM performance audit
- Full container analysis
- All 14 check categories
- Detailed report with remediation plan

Quick Tips

Container under 50KB compressed is healthy; over 100KB needs attention
More than 30 active tags usually indicates cleanup opportunity
Custom HTML tags are the biggest performance risk - audit these first
All Pages triggers are the most common source of waste
dataLayer pushes before GTM loads are queued and processed in order

Adoption

kanopi/gtm-performance-audit

$ install --global

Security Scan Results

SKILL.md

GTM Performance Audit

Philosophy

Core Beliefs

Why GTM Audits Matter

When This Skill Activates

Decision Framework

What Data Is Available?

What's the Goal?

Decision Tree

Setup

Phase 1: Baseline Performance

Step 1: Read the Collector Script

Step 2: Navigate and Run the Collector

Step 3: Supplemental Data (if needed)

Phase 2: GTM Container Analysis

If Container JSON Available

If Only Container ID Available

DataLayer Analysis

Phase 3: Issue Detection

1. Synchronous Script Loading

2. Blocking Tags

3. Large Payloads

4. Main Thread Blocking

5. Missing Conditional Firing

6. Trigger Optimization

7. Duplicate Tags

8. Orphaned Tags

9. Expensive Variables

10. Missing Async Attribute

11. Custom HTML Best Practices

12. Server-Side Candidates

13. Consent Mode Gaps

14. Tag Firing Order

Phase 4: Report Generation

Executive Summary

Detailed Findings

Tag Inventory Table

Network Waterfall

Implementation Checklist

Required MCP Integration

CMS-Specific GTM Patterns

Drupal

WordPress

Guardrails

Integration with /audit-gtm Command

Quick Tips

Related Skills

kanopi/strategist-site-audit

kanopi/story-point-estimator

kanopi/qa-review

kanopi/project-heartbeat

kanopi/gtm-performance-audit

$ install --global

Security Scan Results

SKILL.md

GTM Performance Audit

Philosophy

Core Beliefs

Why GTM Audits Matter

When This Skill Activates

Decision Framework

What Data Is Available?

What's the Goal?

Decision Tree

Setup

Phase 1: Baseline Performance

Step 1: Read the Collector Script

Step 2: Navigate and Run the Collector

Step 3: Supplemental Data (if needed)

Phase 2: GTM Container Analysis

If Container JSON Available

If Only Container ID Available

DataLayer Analysis

Phase 3: Issue Detection

1. Synchronous Script Loading