Opinionated agent skills for integrating the Trackbridge SDK — Google Ads + GA4 dual-send conversion tracking — loaded on demand by your LLM coding assistant.
When Trackbridge is mentioned in your project (a @trackbridge/browser or @trackbridge/server import, an env var like GOOGLE_ADS_DEVELOPER_TOKEN, a question like "wire up a purchase conversion"), your assistant loads the matching skill and follows it correctly the first time — instead of rediscovering the SDK's API and conventions from the README every conversation.
A skills loader CLI is forthcoming for the Trackbridge ecosystem. Once shipped, add an entry to your project's skills-lock.json:
{
"version": 1,
"skills": {
"trackbridge": {
"source": "trackbridge/skills",
"sourceType": "github"
}
}
}…and run the loader's update command. All five skills (the router + four specifics) install together.
Manual install (works today). Until the loader ships, vendor the skills directly:
git clone https://github.com/trackbridge/skills.git .claude/trackbridge-skillsThen point your assistant at the <skill-folder>/SKILL.md files. The exact discovery path depends on your assistant — Claude Code reads .claude/skills/, others vary.
| Skill | Triggers when… |
|---|---|
trackbridge |
the user mentions Trackbridge, @trackbridge/browser, or @trackbridge/server. Provides a high-level overview, three non-negotiable invariants, and routes to the specific skill for the task. |
trackbridge-setup |
the user is adding Trackbridge for the first time — installing both packages, creating the browser + server tracker modules, declaring env vars, verifying the install. |
trackbridge-conversions |
the user is firing a trackConversion call — placing it in checkout success and the matching server webhook, using a shared transactionId, forwarding click identifiers (gclid/gbraid/wbraid). |
trackbridge-ads-oauth |
the user is setting up the Google Ads API for the server tracker — developer token, OAuth client, refresh token via Playground, mapping conversionActions resource names. |
trackbridge-consent |
the user is integrating with a CMP (Cookiebot, OneTrust, Iubenda, custom) for Consent Mode v2 — wiring updateConsent from grant/deny callbacks, handling deferred cookie persistence. |
Skills are content-addressed via computedHash in your skills-lock.json. Update by re-running your loader's update command — there's no semver, no changelog, no npm registry. Tags exist (v0.1.0, etc.) for human bookmarking only; the loader consumes main.
See CONTRIBUTING.md for skill folder conventions, frontmatter rules, and the validator that runs in CI.
github.com/trackbridge/sdk— the SDK these skills cover.trackbridge.dev— product site and end-user documentation.
MIT.