Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    GTM Tracking Auditor

    2

    Agency-grade Google Tag Manager audits: broken tags, GA4 and Ads gaps, Consent Mode v2, server-side checks, and reviewed fixes applied in a workspace.

    $2

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    GTM Tracking Auditor

    GTM Tracking Auditor

    Example session with this skill installed

    Audit the live container GTM-ABC1234 of example-shop.com, tell me why GA4 purchase events dropped since mid-September, and prepare the fix in a workspace. Do not publish.

    • Read your context and instructions
    • Compiled the gtm tracking auditor

    Summary

    GA4 purchase events on example-shop.com stopped on 16 September, when version 41 "Updated page_view" went live. The event tag itself was not touched: the trigger it uses, CE - purchase, gained a condition {{Page Hostname}} equals www.example-shop.com the day the checkout moved to checkout.example-shop.com. The condition is false on every order page, so the tag never fires. Google Ads conversions kept working because they use a different trigger. Fix prepared in workspace

    Fix - purchase trigger - 2026-10-13 (3 changes), not published.

    Audited: live version 44 (89 tags, 57 triggers, 48 variables) and the fix workspace. Audit language=en; report files saved under the reports directory.

    What loses data now (high)

    EntityIssueSinceEffect
    GA4 - purchase - Site (tag 112)trigger condition never true on the checkout hostnamev41, 16 Sepno purchase in GA4; revenue reports empty
    GA4 - begin_checkout - Site (tag 98)same trigger family, same conditionv41, 16 Sepcheckout funnel broken from step 1
    META - Purchase (tag 130)fires from the dataLayer event, unaffected, but event_id is not shared with the server CAPI tagv37Meta counts each purchase twice

    At risk (medium)

    • 6 marketing tags fire on All Pages without a consent condition (META - PageView, LinkedIn - Insight, 4 remarketing). Consent default is set on Consent Initialization with wait_for_update 0: tags decide before the banner answers. Recommended 500.
    • Criteo - OneTag (HTML) and Criteo - OneTag (template) both active: duplicate firing.
    • Conversion Linker present, but cross-domain linking lists only www.example-shop.com: add checkout.example-shop.com.

    Hygiene (low)

    18 orphan triggers, 6 unused variables, 3 paused seasonal tags (kept on purpose, dated in the report). Proposed as a separate cleanup workspace after the fixes are live.

    Change set applied in workspace "Fix - purchase trigger - 2026-10-13"

    1. update_trigger 57 CE - purchase: hostname condition replaced with {{Page Hostname}} ends with example-shop.com.
    2. update_trigger 58 CE - begin_checkout: same.
    3. create_variable Unique Event ID (custom JS) and update_tag 130 META - Purchase to send it as event_id; the server CAPI tag already reads event_id from the incoming event.

    gtm_diff live → workspace: 1 added, 0 removed, 3 modified (fields: filter on 57 and 58, parameter on 130). Nothing else changed.

    Next steps (for you)

    1. Preview on a test order across www and checkout: purchase fires once in GA4 DebugView with transaction_id, value, currency, items; Meta Pixel Helper shows one Purchase with an event_id.
    2. Create version "Fix purchase trigger and Meta event_id", publish from the GTM UI.
    3. Outside GTM: add checkout.example-shop.com to GA4 cross-domain settings and to the Conversion Linker domains; then I can delete the workspace.

    Not covered: the site's own gtag snippet, the server container (audit it separately), GA4 property settings.

    (Demo container and figures are illustrative.)

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Turns Google Tag Manager into the audits and fixes a tracking team actually delivers, on top of the free, open-source mcp-gtm-audit MCP server (read-only by default, opt-in write mode, never publishes).

    What it does

    • Health audit of the live version and of pending workspaces: tags that never fire, dead variable references, Google tags without Measurement ID, GA4 events without name or parameters, Custom HTML risks, duplicated pixels, missing Consent Mode checks, orphan triggers, unused variables. Severity by data impact, with the known false positives and the fix for each code.
    • Tracking validation: every GA4 event per Measurement ID with its parameters, ecommerce parameter gaps (purchase, begin_checkout, add_to_cart, view_item, generate_lead), Google Ads conversions and Conversion Linker, Meta pixels and CAPI deduplication, server-side detection.
    • Consent Mode v2 checklist per tag: fires before consent, after, or gated.
    • Server-side review and migration: the two classic breakages (hostname filters on event data the client never sends, Google tag firing after the first page_view) and a step-by-step partial rollout with a lookup table per hostname.
    • Pre-publish review: diff of a workspace against the live version with the fields that changed and their before/after values.
    • Change sets: with write mode, a dedicated workspace, entities created in dependency order from documented JSON shapes, diff pasted as the review document; without write mode, an importable container JSON for Admin > Import container. The user reviews in Preview, creates the version and publishes.
    • Diagnosis table: symptom → usual cause → where to look, including two defects masking each other.
    • Client-ready tracking audit: structure, honest framing, impersonal voice, actions outside GTM listed separately. A bundled script turns the audit JSON into Markdown tables.

    Ground rules the skill enforces

    Audit what is live first, then what is about to be. Severity is about data, not tidiness. Static rules say what is configured, not what fires: runtime confirmation is named when it matters. One change set, one named workspace, never the Default Workspace, never publish. Read the entity shape before writing. Respect the API quota. Say what was not checked.

    Where it comes from

    Built from real agency engagements: retail container migrations, utilities moving to server-side tagging, B2B sites whose events vanished on a date nobody noticed. Works in Claude Code, Cursor, Codex CLI and any agent that reads SKILL.md and connects to MCP servers.

    Requirements

    The mcp-gtm-audit MCP server (https://github.com/acamolese/gtm-audit-mcp, MIT, pipx install mcp-gtm-audit) connected to the agent, with Tag Manager access to the container. Write mode needs the server started with GTM_ALLOW_WRITE=1. No credentials in the skill. One bundled Python standard-library script that reads the audit JSON locally: no network, no environment variables.

    Static analysis of the container: it sees what is configured, not what fires at runtime (dataLayer, CMP, race conditions), so the skill names what still needs confirming in GTM Preview or GA4 DebugView. Read-only unless the MCP server is started in write mode; even then it only changes workspaces and never creates a version or publishes. The site's own gtag or pixel snippets, GA4 property settings and Google Ads settings are outside the container and are listed as actions for the user. The Tag Manager API meters writes tightly: change sets are a handful of entities, not a container rebuild.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Requires

    Google Account
    Gtm Audit MCP
    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Allowed Hosts

    tagmanager.google.com

    Terminal: runs the bundled scripts/audit_summary.py (Python standard library only) on the audit JSON the MCP server wrote locally. Read/Write files: reads that JSON and writes the audit document the user asked for, where the user says. No network access from the skill itself: all Tag Manager data goes through the mcp-gtm-audit MCP server, which the user installs and authorizes separately (read-only OAuth scope by default, container-edit scope only if they opt into write mode). No environment variables, no credentials in the package.

    Listed1 month ago
    Updated22 days ago

    What's inside

    Frequently Asked Questions

    This skill uses a specialized MCP (Model Context Protocol) server to bridge your AI agent with the live GTM API, allowing it to read your actual container JSON, identify schema errors, and draft real-time fixes. Proprietary prompts alone can't 'see' your workspace or verify if triggers are correctly configured.

    Any agent that reads SKILL.md and can call MCP tools: Claude Code, Claude Desktop, Cursor, Codex CLI and others. It needs the free mcp-gtm-audit MCP server installed (pipx install mcp-gtm-audit), authorized once with your Google account through OAuth (read-only scope by default), and Python 3.10+ for the bundled summary script.

    No. With write mode on, the agent creates or changes tags, triggers and variables inside a dedicated workspace, never the Default Workspace, and the MCP server has no publish and no version tool: the publish OAuth scope is never requested. Every change set ends with the same step: review in GTM Preview, create the version and publish from the GTM UI yourself. Without write mode, the agent delivers the same changes as an importable container JSON.

    It checks the facts: whether a consent default command runs on Consent Initialization, how long tags wait for the CMP (wait_for_update), whether ads_data_redaction and url_passthrough are set, which tags carry Additional Consent Checks, and which marketing tags fire on All Pages without a consent condition. The report lists every tag as "fires before consent", "after consent" or "gated". The legal reading of those facts stays with you or your DPO.

    A SKILL.md with the routing between eight playbooks (health audit, tracking validation, Consent Mode v2, server-side review and migration, pre-publish review, change set, diagnosis, client report), seven reference files with the method behind each one (tool cheat sheet, audit rules with false positives and fixes, GA4/Ads/Meta/consent/server-side checks, JSON shapes for tags, triggers and variables, a symptom-to-cause diagnosis table, the client report structure), and one Python script that turns the audit JSON into Markdown tables. The MCP server itself is a separate, free, open-source project.

    New in Marketing & SEO