Apply Post for Me's conversion-tracking conventions to any change that touches user/team lifecycle, billing, signup, the Stripe webhook, PostHog/Meta/Google integrations, or anything under `app/tracking/`. Use this skill whenever adding/modifying analytics events, working with PostHog identify/group calls, handling Stripe webhook events, building checkout or billing flows, wiring ad attribution, or adding any new user-lifecycle moment that might be worth tracking. Reach for it any time someone says "track this", "add a posthog event", "log this conversion", "do we capture X anywhere", or "should we be tracking…" — even if the word "conversion" isn't said. Default to firing when in doubt; under-tracking is the failure mode this skill exists to prevent.
Instalación
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Apply Post for Me's conversion-tracking conventions to any change that touches user/team lifecycle, billing, signup, the Stripe webhook, PostHog/Meta/Google integrations, or anything under `app/tracking/`. Use this skill whenever adding/modifying analytics events, working with PostHog identify/group calls, handling Stripe webhook events, building checkout or billing flows, wiring ad attribution, or adding any new user-lifecycle moment that might be worth tracking. Reach for it any time someone says "track this", "add a posthog event", "log this conversion", "do we capture X anywhere", or "should we be tracking…" — even if the word "conversion" isn't said. Default to firing when in doubt; under-tracking is the failure mode this skill exists to prevent.
Conversion tracking
Post for Me sources every paid-lifecycle event from Stripe webhooks (not in-app actions) and ships them to PostHog, which forwards conversion events to Meta Ads and Google Ads via destinations. The whole funnel — discovery → signup → conversion → upgrades → cancellation — runs through this pipeline.
There are two classes of tracked events, and the distinction governs where they fire:
Paid-lifecycle events (customer_converted, subscription_*) — sourced from the Stripe webhook, never from in-app handlers. The webhook is the honest signal (it fires regardless of source: UI, API, support agent, dunning). Live in subscription-lifecycle-tracking.ts.
User-lifecycle events (user_signed_up, team_created, project_created/_deleted, team_member_invited/_joined/_removed) — sourced from in-app route actions, because that's where the user intent actually happens (there's no webhook for "user created a project"). Consolidated in tracking/.server/lifecycle-tracking.ts.
Tracking code is centralized in app/tracking/ (browser) and app/tracking/.server/ (server) in both dashboard/ and marketing/. Route-specific firing happens in route files but always imports from the namespace. Each app's tracking/README.md is the tour map — read it before adding things.
The attribution model — never violate this
PostHog person = the user.distinct_id = Supabase auth user id.
Browser: posthog.identify(user.id, …).
Server (webhook, cron): distinctId = teams.created_by (same user id).
PostHog group team = the billing entity, keyed by teams.id.
Browser: posthog.group("team", team.id, …).
Server: setTeamGroupProperties(team.id, …)andteamId: team.id on captureServerEvent.
PostHog group project = a project, keyed by projects.id. A user acts on behalf of one project at a time.
Browser: posthog.group("project", project.id, …) (fires only on project-scoped routes).
Server: setProjectGroupProperties(project.id, …)andprojectId: project.id on captureServerEvent. Project-scoped events attach both the team and project groups.
Two group types are defined (team, project); PostHog allows up to 5. Don't add a third without a reason.
All billing reporting is group-aggregated (unique teams), not unique users. A user can be in many teams — group analytics handles it naturally.
PostHog has no person-less events. You always pass a distinctId. For server-sourced billing events that's the team owner. The person is plumbing; the team group is the reporting unit.
Hard rules for any new tracked event
Paid-lifecycle events come from the Stripe webhook, never from in-app button handlers. The webhook fires regardless of source — UI, API, automated flow, support agent in Stripe — and that's what makes the signal honest. The webhook lives at dashboard/app/routes/api.stripe.webhook/; analytics for subscription events lives in .server/subscription-lifecycle-tracking.ts adjacent to it. User-lifecycle events (signup, team/project/membership) are the exception: they fire from the in-app route action that performs the mutation, via tracking/.server/lifecycle-tracking.ts (fire-and-forget). There's no webhook for them, and the in-app action is the honest source.
Dedupe deterministically. Use deterministicUuid("<event_name>:<stable_entity_id>") as the event's uuid — derived from IDs we own (user.id, subscription.id, team.id, …), never a random UUID. Caveat (verified against the project): PostHog does not drop duplicate-uuid rows at ingestion or query time — its uuid dedup is an eventual ClickHouse merge, not an upsert. So re-emitting an event adds a duplicate row that count() sees; a re-run is not silently idempotent. The deterministic uuid's real payoff is that duplicates are collapsible after the fact: count(DISTINCT uuid) (or DISTINCT <entity_id>) always yields the true count. Practical rule: the live webhook fires once per event so this rarely bites; for the backfill, prefer not to re-emit (run once / resume after a crash) and always analyze on DISTINCT uuid.
Stamp the source system's time. Webhook events: new Date(event.created * 1000). Real-time events (signup, cron decisions): default to now. Never let PostHog use ingestion time — retries and resends would land at the wrong moment.
Attach the team group when the event is about a team or its billing (teamId: team.id on captureServerEvent). Also attach the project group (projectId: project.id) when the event is about a specific project. Project-scoped events carry both.
Wrap analytics in try/catch at the call site.captureServerEvent already swallows errors, but the helper that prepares the call (resolves team, reads metadata, computes properties) can still throw — keep it from taking down the surrounding webhook/cron handler.
system_triggered: true marks automation; absence implies human-driven. Only set the flag on events fired by the system (cron, scheduled tasks). Don't require it on every event — the failure mode would be silent mislabeling.
No cross-sibling imports.dashboard/ and trigger/ keep vendored copies of tracking/.server/posthog.ts. Keep them in sync by hand. Don't introduce a shared package or workspace alias.
Pick the right property home — person / group / event:
Person properties: identity / matching data (email, first_name, last_name, gclid, fbclid, fbc, fbp, utm_*). Set via posthog.identify browser-side. Meta/Google destinations primarily read these.
Group properties: current team state (subscription_status, is_active, plan_post_limit, cancel_at_period_end, member_count, project_count). Refresh on every relevant event so "is this team active?" / "how big is it?" is self-healing. groupIdentifymerges, so refreshTeamShape() (counts) and the subscription tracker (billing state) update disjoint keys without clobbering each other.
These fire from in-app route actions via tracking/.server/lifecycle-tracking.ts. Use its helpers (trackTeamCreated, trackProjectCreated, trackProjectDeleted, trackTeamMemberInvited, trackTeamMemberJoined, trackTeamMemberRemoved, refreshTeamShape) rather than calling captureServerEvent from the route directly — that's the user-lifecycle analog of subscription-lifecycle-tracking.ts. Each helper is best-effort and is invoked fire-and-forget (void helper(...).catch(...)) so analytics can't slow or break the action.
role is derived, never stored. There's no role column on team_users. The team's created_by is the owner; everyone else is a member — roleForTeam(team.created_by, userId). If a real roles system lands later, replace the derivation everywhere, don't add ad-hoc role columns.
Signup splits invited vs organic. Every new user gets a personal auto-team + default project from DB triggers. An invited user (user_metadata.source.type === "invite") is attributed to the team they joined (role: member); an organic user to their personal team (role: owner). The live path skips invited users' throwaway personal teams for team_created/project_created; the backfill replays the whole table (dedupe keeps it idempotent).
Joined = seat activation.team_member_joined fires when an existing user is added (immediate) or an invited new user first authenticates — not at invite time. The team creator's own seat is covered by team_created, not a join event.
Counts are self-healing.refreshTeamShape() recomputes member_count/project_count after every membership/project change.
Auto-created default projects count as user-driven. We're moving away from auto-creating projects; until then, treat the trigger-created default project as if the user made it (fire project_created for it) so the activation signal stays continuous.
Plans are ordered by post limit, not tier index
A higher plan_post_limit is a higher plan. When emitting upgrade/downgrade direction, compare to_post_limit > from_post_limit. Do not introduce tier indices or new ordinal columns — plan_post_limit is monotonic across PRICING_TIERS and is the natural ordering.
Carrying browser context into server events
Server-side events fired from the Stripe webhook have no browser context — no $current_url, no $raw_user_agent, no gclid from the active URL. The dashboard captures these at checkout-initiation (in _protected.$teamId.billing._index/route.loader.ts) and stamps them into subscription_data.metadata. The webhook then reads subscription.metadata and stuffs them onto customer_converted / subscription_reactivated. If you add a new conversion-style event that ad destinations should pick up, source the ad-attribution props the same way — from subscription.metadata via adAttributionProps() in subscription-lifecycle-tracking.ts.
PostHog identifier mount (team + project groups) → routes/_protected.$teamId/route.component.tsx
Backfill (backfill/, local-only, gitignored) replays history into PostHog. Tasks: group-state, signups, teams-created, projects, team-members, conversions, plan-changes. Add a task + register it in run.ts whenever you add a backfillable event. See backfill/README.md.
If you're adding a new subscription-lifecycle event, the analytics belongs in subscription-lifecycle-tracking.ts. If you're adding a new user-lifecycle event (team/project/membership/signup), it belongs in lifecycle-tracking.ts. Either way, don't sprinkle captureServerEvent calls across unrelated route handlers — the consolidation is intentional.
Checklist for adding a new tracked event
Event name is snake_case, action-oriented (customer_converted, not userConvertedToCustomer).
distinctId resolves to a real auth.users id (browser: user.id; server: teams.created_by).
teamId is passed when the event is team-scoped; projectId too when it's project-scoped.
Subscription event → subscription-lifecycle-tracking.ts; user-lifecycle (team/project/membership/signup) → lifecycle-tracking.ts. Not sprinkled in the route handler.
dedupeKey is deterministicUuid("<event_name>:<stable_entity_id>").
timestamp reflects when the event occurred, not when it was ingested.
Properties include team_id, stripe_customer_id (if billing), subscription_id (if subscription-related).
If conversion-style, also include price, currency, and ad attribution via adAttributionProps(subscription.metadata).
Call site is try {…} catch {…} so analytics failures can't take down the handler.
system_triggered: true is set only if the event was fired by automation (cron/scheduled task).
references/event-catalog.md updated as part of the same PR.
If the event is backfillable from existing data, add/extend a backfill/ task and register it in run.ts.
If the event should reach Meta/Google: confirm the destination's "Match events" filter includes it, and check references/destination-mappings.md for what properties each destination reads.
Anti-patterns — refuse these
Firing paid-lifecycle events from a browser button handler. (Misses support-driven changes, dunning cancellations, automated upgrades.)
Setting timestamp to new Date() on a webhook handler. (A retried 5-minute-old Stripe event would land at the wrong time.)
Reading person properties (gclid, fbclid) directly from PostHog inside a server handler — they're not in scope there. Source from subscription.metadata (set at checkout).
Treating trialing as a paid conversion if/when trials are introduced. Current code does because we don't offer trials; add a separate trial_started event and redefine customer_converted as first paid subscription when that changes.
Using a random uuid as the dedupe key. PostHog dedup requires determinism; randoms allow doubles.
Cross-sibling imports — vendor the helper into each sibling instead.
Adding new tier-ordinal columns or indices. Use plan_post_limit for ordering.
Inventing a role column / table to populate the role property. It's derived from teams.created_by (roleForTeam). Schema changes for analytics are out of scope here.
Firing team_member_joined at invite time. It's a seat-activation signal (existing user added, or invited user's first auth) — invite time is team_member_invited.
When to escalate to the user
Ask before doing — these are not defaults:
Adding a browser-side fbq('track', …) or gtag('event', …) for conversions. (We deliberately fire conversions only via PostHog → destinations, no browser dedup work.)
Adding new PostHog destinations or modifying their event-name / property mappings. (Lives in PostHog UI, not the codebase, but affects what properties events must carry.)
Adding historical_migration: true semantics or backdating to events older than ~1 year.
Introducing a non-Stripe source for a paid-lifecycle signal.
References
references/event-catalog.md — canonical list of currently tracked events with the properties each carries. Update with every PR that adds or modifies an event.
references/destination-mappings.md — what each PostHog destination reads (Meta CAPI, Google Ads). Useful when adding a new event that should reach an ad platform.
