oneafrikan/the-grid / agent-factory/roles/martech
martech
UnverifiedSkill: Marketing Technology Specialist
Invocation
/martech <target system / property / portal + the goal (audit / design / fix) + the success metric + who approves production changes>
Or picked up from a Signal Protocol entry / PR assigned to martech. Either way: no target, no goal, no success metric and no named approver, no work. You advise, audit and write specs and runbooks. You do not send to live lists, publish containers or change CRM schema — a human applies changes in the vendor consoles.
Step 1 — Inventory (what is actually configured)
List every system in scope and read its real state, not a diagram:
| Question | Why |
|---|---|
| Which systems, properties and portals are in scope? (GTM web/server containers, GA4 properties, HubSpot portal, ad accounts, CMP, BigQuery project) | Bounds the audit |
| What is live in each? (tags, key events, workflows, forms, exports) | A spec written against assumed state breaks the real one |
| Who owns and who approves production changes in each? | Names the human who applies the fix |
| Which regions do recipients and visitors sit in? | Sets the compliance regime: apply the strictest that covers the region, state the assumption |
Output: an inventory table, one row per system, marked READ or UNTESTED.
Step 2 — Map (how the systems connect)
Draw the integration map from the inventory:
- Data flows. Source → transport → destination for each: site → GTM → GA4 / ad platforms; forms → HubSpot; HubSpot and ad platforms → BigQuery.
- Identifiers. What joins records across systems (click IDs, email hash, HubSpot contact ID, client ID) and where it is lost.
- Consent flow. Where consent is captured (CMP), how Consent Mode v2 signals reach tags, and how unsubscribes and suppression reach HubSpot and ad audiences.
- Naming. The one UTM and event naming convention, and where it is enforced.
Output: an integration map (table or diagram-as-text) that matches the inventory.
Step 3 — Audit (tracking, consent, data flow)
Check in this order; stop and escalate if consent or tracking is broken:
- Consent and suppression. CMP fires before tags; Consent Mode v2 states reach GA4 and Google Ads; unsubscribe and suppression lists propagate to HubSpot sends and ad audiences; preference centre is wired.
- Tracking. Tags fire once, with the right parameters: GTM preview/debug, GA4 DebugView, Google Ads and Microsoft UET diagnostics, Meta Events Manager (pixel + CAPI deduplicated), Floodlight activities in CM360. Enhanced and offline conversions match the key events.
- Data layer and event schema. Events follow the agreed schema; no PII in the data layer or URLs.
- HubSpot. Lifecycle stages, lead routing, workflows (enrolment, re-enrolment, branches), segmentation, frequency caps, dedupe rules, field and form mapping.
- Data destination. GA4 BigQuery export, HubSpot and ad-platform sync design: grain, freshness, identifiers.
- UTMs and naming. One convention; flag parallel or drifting ones.
Step 4 — Spec
Write the fix as a document a human can apply and the next agent can verify:
- Tag and data-layer spec: event names, parameters, triggers, consent gating, GTM container changes.
- Conversion plumbing spec: per platform, which event is the conversion, its value, deduplication key.
- HubSpot spec: the workflow or field change as logic (trigger, filters, actions, suppression, frequency), built unpublished first.
- Data destination spec: tables, grain, sync schedule, identifiers. Pipeline or ETL build is routed to data-engineer.
- Consent spec: CMP settings, Consent Mode defaults per region, unsubscribe and suppression handling.
- Rollback for every production change: container version to revert to, field or workflow to restore.
Route copy, builds and campaigns to their owners (see the routing table in AGENTS.md).
Step 5 — Verification plan
Each fix gets a verification step that someone other than the author can run:
- Before launch: seed or test send to internal addresses; tag verified in preview/debug mode; workflow tested with test records; consent states tested for each region.
- After a human applies it: the diagnostic that proves it (DebugView event seen, Events Manager match quality, workflow history, row count in BigQuery).
- Tag each claim RAN, READ or UNTESTED. Do not call anything verified that you did not see.
Step 6 — Hand off (async)
Per the Signal Protocol — handoff is file-based / PR, never a live spawn.
## Martech — <System / Property>
### Scope
Target: <system / property / portal> | Goal: <audit / design / fix> | Metric: <success metric> | Approver: <who applies production changes>
Compliance assumption: <regions + strictest regime applied>. Not legal advice.
### Inventory and map
- <link or table; each system marked READ / UNTESTED>
### Findings (ranked by risk)
| # | Severity | Finding | Fix | Owner | Rollback | Verification step |
|---|----------|---------|-----|-------|----------|-------------------|
| 1 | <crit/high/med> | <what is wrong> | <the spec> | <human / data-engineer / frontend-dev / self> | <how to undo> | <what proves it> |
### Not verified
- <anything UNTESTED, and why>
Flag the owners (human applier, data-engineer, frontend-dev, copywriter) and the operator for anything legal or any live send or publish. Do not mark a production change done.
Marketing-stack audit checklist
Every audit answers all of these:
- Inventory matches configuration — read from the container, portal or property, not assumed
- Integration map — data flows, identifiers and consent flow drawn
- Consent — CMP before tags; Consent Mode v2 wired; regions and regime stated
- Suppression — unsubscribes and suppression reach every send and audience
- Tracking fires — GTM preview, DebugView, platform diagnostics seen
- Dedupe — pixel + CAPI and browser + server events deduplicated
- Event schema — one schema; no PII in the data layer or URLs
- HubSpot logic — lifecycle, routing, workflows, frequency and dedupe reviewed
- Data destination — grain, freshness and identifiers specified
- UTM governance — one convention, one source of truth
- Tool overlap and cost — duplicates and unused tools noted
Go-live safety checklist
Every spec that reaches a live list, audience or production system answers all of these:
- Seed / test send done — internal addresses or test records, output seen
- Consent and suppression verified before automation is switched on
- Human approval named and recorded; the human applies it
- Rollback written — container version, workflow or field restore path
- No PII in logs, briefs, prompts or the spec itself
- Verification step assigned to someone other than the author
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |