Documentation version: 2026-10-23.6
Canonical: https://docs.creator.gg/guides/growth-routine.md

# Daily growth habits and shared routines

## Start with today
When the user asks “What do we need to do today?”, read workspace_status, intelligence_get and growth_today_get. Use the current goal and agreed routine. If the routine is not enabled, its habits are starting suggestions. Agree or edit cadence, ownership, channel, first due date and what counts as done, then save with growth_routine_update and the current revision. This saves commitments, not authorization to send, publish or create a schedule. A missing channel is a preparation/setup task, not a failed public action.

Start with daily useful social content and audience engagement. Weekly batching makes daily distribution easier. Useful segmented email can start every fourteen days, meaningful product updates every 2–4 weeks, and regular CTA/conversion work weekly. Customer conversations, lifecycle checks and relevant seasonal campaigns can be added. Adapt to the actual business, objective, available audience and resources. Do not force an acquisition routine onto a retention-only goal. No-reply days can complete a useful engagement review; never invent replies to earn completion.

## Claim and act
Read growth_today_get. It returns stable occurrence IDs, deadlines, owners, completion rules, evidence and active claims. Claim one with growth_task_claim and a stable requestId. Retain its claimToken. Claims have a 5–60 minute lease; another agent cannot claim the same task during the lease. Keep claimToken private to this connection. Before any external mutation, preserve the existing operation receipt and approved content/audience. A lease expiry is never permission to repeat an uncertain send or publication. Reconcile provider state first. Claims coordinate these routine occurrences; they are not a global lock on all Creator Agents actions.

Prepare drafts and assets, resolve needed human input, then execute only within the user's existing authorization. Scheduled work is preparation for the due day. It does not complete a publication habit. An agent does not have to be running for dates to become due: occurrences are derived from the saved cadence in the business timezone. This system creates no new agent runtime, cron or autonomous publisher. Optional morning automation belongs to the user's supported host and requires their opt-in.

## Record evidence
Use growth_task_record with taskId, claimToken, note, state and requestId. States are prepared, scheduled, blocked and done. For social/email completion, supply the native post/campaign ID. Creator Platform reads the authoritative provider record and requires confirmed status plus a publication/send event date. A scheduled/pending/unknown state or missing event date returns verification_unavailable. Missing proof does not become verified merely because a tool call returned completed. Campaign sent evidence is separate from recipient delivery evidence.

Without a native resource ID, completed work is labeled agent_reported. Human dashboard entries are user_reported. Both count toward agreed habits, with their source always visible. Report manual/external work honestly and keep the note concrete. Do not mark work done solely for opening the dashboard, generating drafts, claiming a task or counting API calls. The same provider event cannot complete multiple occurrences. Per-channel social habits use an accountId returned by social_accounts_list. Use separate habits for daily channel commitments.

Provider publications and eligible campaign sends can also be reconciled from native activity. Reads are bounded, and coverage is returned. Missing timestamps or an unsupported envelope stay unverified. Social reconciliation covers up to 20 recent publications; queue reading covers up to 100 records and reports partial coverage. No public action is submitted by reconciliation. Record an exact ID for evidence outside the bounded window. Prepared/scheduled work is never silently converted into completion without provider evidence.

## Dates, edits and streaks
Cadences use calendar dates and an IANA timezone, rather than elapsed 24-hour intervals. An initial routine starts today; edits start tomorrow, preserving previous versions, deadlines and evidence. Periodic dates stay anchored. A changed cadence or new habit cannot be backdated. Edits made before tomorrow starts replace that pending version without erasing today's history. Use the actual occurrence ID after refresh.

The calendar measures commitments completed by their due day. Extra easy activity cannot compensate for a missed commitment. Late periodic work can clear the overdue task but does not repair the missed calendar day. Today's unfinished work does not break yesterday's streak before the day ends. Current and best streaks, per-habit streaks and complete-day milestones retain the full saved history through daily summaries. Completion rate and the visible matrix show the selected recent calendar window. All task evidence and routine versions remain stored. Free-to-native linking moves a routine and its history when the destination has no routine. If both have routines, linking returns routine_link_conflict so the customer can resolve ownership without a silent overwrite.

## Metrics and review
Read growth_metrics_get for dated observations, evidence sources, connected account IDs and queue coverage. Native CRM contacts are labeled contacts. They include customers and other records; do not call them qualified leads or active product users. Use growth_metrics_record for explicitly sourced external observations with a stable definition and date. Those observations remain reported. Changes to the definition require reconciliation rather than a misleading continuous series. Missing metrics are unavailable, not zero.

The dashboard charts show weekly completion, dated results and saved plan scenarios. Illustrative signup/activation assumptions are separated from measured results; a streak does not establish a growth forecast. Milestones represent review readiness and accumulated evidence, not guaranteed users or locked basic capabilities. Ask for a weekly review: what happened, what the audience taught us, which bottleneck matters, and what to change next.