conversion-tracking helps an agent build tracking that provably works, or find why existing tracking does not. It separates lead generation, where the page has to signal success, from ecommerce, where the order record is the source of truth, and picks the checks that fit each.
Setup and audit flows both start from three facts: what counts as a conversion, where it must land, and what tool runs the moment. A script turns a form tool and an ad platform into one GTM container file you import and review in Preview before anything goes live.
What You Get
- A setup flow that fires on confirmed success and not on the submit click
- An audit flow routed by symptom, such as doubled conversions or zero conversions
- A GTM container import file for a chosen tool and ad platform, built with your own IDs
- Verification steps with a named check and expected result for each part the agent cannot reach
- A handover note with IDs, event names, expected loss for your traffic and what is likely to break later
Common Use Cases
- Tracking form submissions as Google Ads or Meta conversions
- Finding why GA4 and Google Ads conversion counts differ
- Fixing doubled conversions from a pixel and a server event
- Choosing between browser tags, Enhanced Conversions and server-side sending
What Makes It Unique
It treats a firing tag and an arriving conversion as two separate claims, and every finding is labeled observed, reported or inferred. A symptom gets a diagnosis and a named fix before any change. It also compares delivery paths including free and native ones, and discloses the one partner-marked tool in its server-side reference.
Under The Hood
The agent loads the SKILL.md and then the references a task needs. It reads your product marketing context file, fetches the site’s pages to see existing tags, searches your codebase for form handlers and tracking calls, and can read a GTM container or ad account you give it. A real test submission fires a real conversion, so the skill requires the exact URL and a yes first. The build script is Python using the standard library: it checks every ID against its platform format and writes one JSON import file to disk. A check script validates the recipes. It makes no network calls of its own.