Audit any website's GA4 and GTM measurement setup from the outside. No Google account access required.
adwize-audit crawls a public page, parses the GTM container, runs 26 deterministic checks against a checkpoint registry, and produces a scored report (A-E) with actionable findings. An optional LLM agent adds an analyst brief when you provide an OpenAI API key.
| Category | Examples |
|---|---|
| Data Collection | GTM installed and in <head>, single container, dataLayer order, event inventory, ecommerce funnel, recommended event names |
| Data Quality | No duplicate GA4/Ads tags (hardcoded + GTM), event naming convention, custom HTML tag count, third-party script sprawl |
| Privacy | Consent Mode configured and transmitted, PII in event names, PII in network requests, enhanced conversions review, third-party cookies |
| Attribution | Cross-domain linker decoration |
Each check produces a finding (pass / fail / warn / n/a) with a severity level. The final score starts at 100 and deducts points by severity — critical (25), high (12), medium (5), low (2).
# Clone and install
git clone https://github.com/Adwize/adwize-audit.git
cd adwize-audit
uv venv
uv sync --extra dev
uv run playwright install chromium
# Run your first audit
uv run adwize-audit audit scan https://example.comCopy the sample env file (the CLI loads .env on startup):
cp .env.sample .env| Variable | Purpose | Required |
|---|---|---|
OPENAI_API_KEY |
Enables the LLM analyst brief | No — deterministic report works without it |
ADWIZE_MODEL_ANALYST |
Model for the analyst agent (default: gpt-4o) |
No |
ADWIZE_MODEL_SCHEMA_MAINTAINER |
Model for schema discovery (default: gpt-4o-mini) |
No |
ADWIZE_DATABASE_URL |
Postgres URL for shared storage (default: local SQLite) | No |
ADWIZE_SCHEMA_DIR |
Directory for agent-written schema overrides | No |
# Basic scan — terminal output with grade + findings
adwize-audit audit scan https://example.com
# Save a full Markdown report
adwize-audit audit scan https://example.com --report ./audit.md
# Force-fetch a container that consent-gates the GTM snippet
adwize-audit audit scan https://example.com --container GTM-ABC123
# Deterministic only (skip LLM analyst)
adwize-audit audit scan https://example.com --no-analyze
# List and inspect stored runs
adwize-audit audit list
adwize-audit audit show <run_id>Run any command with --help for full options.
See a full example report (Stripe.com) for what the --report output looks like.
core/ Audit engine
collectors/ Playwright crawl + GTM container parser + vendor detection
checks/ 23 deterministic check functions → Finding objects
registry/ Checkpoint definitions (YAML)
schemas/ Knowledge base (GA4 events, CMP patterns, PII, vendors)
models/ Pydantic data models
scoring.py 100-point scoring → grade A–E
agents/ LLM agents (key-gated, optional)
analyst/ Executive summary + deep-dive + next steps
schema_maintainer/ Discovers and classifies unknown GTM tag functions
cli/ Typer CLI with Rich terminal output
storage/ Async SQLAlchemy persistence (SQLite default, Postgres optional)
report/ Markdown report generator
tests/ pytest suite
uv sync --extra dev
uv run playwright install chromium
# Run tests
uv run pytest
# Lint
uv run ruff check .
uv run ruff format --check .See CONTRIBUTING.md for development setup, coding standards, and how to submit changes.