| name | marketing-automation-blueprint |
| description | Design or critically review a complete autonomous marketing operating system across truth-bound positioning, audiences and consent, owned/earned/paid channels, creative and content, SEO/AEO, stores, lifecycle messaging, referrals, cross-promotion, creators, community, partnerships, reputation, attribution, incrementality, budget allocation, experiments, and autonomous maintenance. Use when the primary artifact is a multi-channel marketing control plane or full go-to-market operating blueprint. For one campaign, official social-account operating model, referral, listing, notification, review/feedback loop, or distribution package, use the specialist procedure. |
Marketing Automation Blueprint
Design the complete relevant global marketing capability portfolio now, then let a bounded autonomous control plane decide which approved content or campaigns to activate, spend on, scale, hold, pause, or retire from current product truth, consent, causal evidence, economics, platform authority, quality, and trust. Delegate exact official social-account execution and live readback to social-media-operations-review.
Objective
Maximize incremental durable user value and contribution after fees, acquisition spend, fraud, refunds, chargebacks, support, infrastructure, creative/rights, and uncertainty—subject to claim truth, consent, law, platform, safety, age, brand, deliverability, reputation, and ruin floors.
Followers, impressions, clicks, CTR, opens, installs, CPI, platform-attributed conversions, platform ROAS, gross revenue, and time spent are diagnostics. None is a complete objective.
Atomic boundary
This skill owns one independently useful artifact: the Marketing Automation Blueprint, including the multi-channel desired-state, creative briefs, and spend operating policy. It consumes product promise, claims, prices, entitlements, offer/grant semantics, a planning Product Program Manifest revision when present, and specialist campaign artifacts; it may translate them for a channel but cannot invent or overwrite them. Exact rendered product media belongs to product-asset-production.
One campaign, listing, referral, notification system, review prompt, positioning decision, or public-signal research question remains with its specialist. Add a new skill only when a distinct recurring user trigger and independently acceptable artifact exists; use references for inseparable capability modules.
For selected official social channels, Marketing owns channel role, audience, message hierarchy, campaign/organic brief, activation envelope, allocation, and causal measurement. social-media-operations-review is the unique operational adapter owner for account authority, publishing/readback, listening, reply/moderation routing, crisis, impersonation, recovery, and shutdown. Marketing consumes its observed evidence rather than maintaining a second social publisher or token ledger.
Agent-first invariant
Assume autonomous production capacity is elastic. Evaluate the full owned, earned, paid, lifecycle, creator, partner, store, referral, cross-promotion, measurement, creative/localization, fraud, privacy, and control-plane portfolio. Build every capability selected by the product, audience, channel mechanism, or explicit program target to a production-shaped scale target now; record why unselected channels are inapplicable. Human effort, missing users, unknown demand, or speculative ROI may not shrink the selected target or leave later automation, localization, SDK, hardening, or scale debt.
Capability selection, construction, campaign activation, and paid spend are different decisions. Every selected channel adapter or specialist operating contract can be complete and dormant. Unknown ROI does not block construction of a selected capability; stale or untrustworthy telemetry blocks unbounded spend. Each live channel starts inside a bounded canary and can scale only within declared authority, consent, claim, budget, causal-evidence, saturation, trust, and safety envelopes. Official social activation uses the Social Media Operating Model's exact-account readback and safety state.
Use explicit shutdown states. dormant-never-activated means no token, API call, tracking, audience export, job, publish, bid, budget, spend, or undeclared data flow. paused-or-withdrawing permits only bounded reconciliation, revocation, billing/readback, correction, and evidence authority—never new audience, creative, delivery, bid, or spend. withdrawn-verified requires provider-observed zero delivery/spend and required data revocation before credentials and jobs can be removed and the adapter becomes dormant. For official social, consume these states from the Social Media Operating Model rather than claiming them from Marketing intent.
Build breadth is cheap; audience attention, trust, rights, platform accounts, external spend, and correlated reputation risk are scarce. Agent-first does not authorize spam, fabricated proof, policy evasion, covert endorsements, review manipulation, low-value scaled content, vulnerability targeting, self-approved spend, or unlawful data use.
Composition contract
When orchestrated, Marketing consumes an immutable planning revision and
remains upstream of the observed-state Product Program Manifest revision:
App/Game Design + commercial truth [+ Product Program Manifest planning revision N]
-> Marketing Automation Blueprint and creative briefs
-> Product Asset Production when exact rendered media is required
-> channel execution
-> Product Program Manifest observed-state revision N+1 when orchestrated
Consume exact product truth and, when orchestrated, a sealed planning manifest
revision containing release targets and stable handoff IDs. Never consume the
later observed-state revision that indexes Marketing. Follow the product artifact envelope:
top-level artifacts use artifactVersion, artifactRevision, and
artifactState and never self-hash; every input names the exact producer
contract through fulfillsHandoffId; a sealed input reference additionally
requires artifactDigest and digestRule: sha256-exact-bytes, while a draft
input reference contains neither. Never invent them or resolve a moving
“latest” revision.
Resource guide
- Read
references/marketing-thesis-and-portfolio.md for objective hierarchy, audience, positioning translation, channel sweep, and system map.
- Read
references/marketing-organic-lifecycle-and-reputation.md for SEO/AEO, content, social channel selection and briefs, creators, community, lifecycle, cross-promotion, update adoption, ratings, and feedback integration; official-account operation routes to social-media-operations-review.
- Read
references/pc-store-launch-marketing.md for a Steam or other PC-store wishlist/demo/festival, creator/press, community, launch-calendar, and first-week communication program.
- Read
references/marketing-paid-and-creative-production.md for paid adapters, creative/localization supply, rights, fatigue, budgets, bids, and channel contracts.
- Read
references/marketing-measurement-risk-and-economics.md for causal measurement, attribution, risk-adjusted value, privacy, fraud, brand safety, and decision gates.
- Read
references/marketing-autonomous-control-plane.md for manifests, role separation, reconciliation, canaries, failure/recovery, proof, and official source routes.
Read all references for a complete marketing system. For a bounded review, load only touched references and state what remains unassessed.
Operating rules
- Consume a versioned product promise, audience modes, claim-proof registry, price/offer/entitlement source, release state, platform/territory availability, and support truth. Missing proof blocks the claim, not construction of the adapter.
- Label
given, observed, assumed, hypothesis, and decision. Separate platform-reported attribution, observed correlation, controlled incrementality, modeled contribution, and strategic judgment.
- Define audience job/intent/lifecycle state, exclusions, consent, age/territory/sensitive-category policy, expected recipient value, message, channel role, and natural frequency before content generation or spend.
- Sweep website/landing, SEO/AEO, store, content, social, community, creator/influencer, PR, partner/affiliate, referral, cross-promotion, lifecycle CRM, reputation, launch, paid search/social/display/video, creative, experimentation, attribution, incrementality, MMM, fraud, brand safety, and operations.
- Give every selected capability a production-shaped target or typed specialist operating contract, API/portal authority record, numeric throughput/spend/failure envelope, verified slices, dormant-state proof, events/cost ledger, policy, canary, rollback/reconciliation, and maintenance loop. Record a mechanism-based reason for each channel excluded after the full sweep. Do not build a competing official-social account, token, publishing, reply, or live-readback system.
- Treat every claim and creative brief as a versioned artifact with provenance,
rights requirements, locale, accessibility, product/release compatibility,
AI/sponsorship disclosure, expiry, and acceptance tests. Marketing owns the
campaign brief, concept, message, variant hypothesis, and selection intent;
product-asset-production owns exact captures and rendered files.
- Consume the product's Durable Investment Contract. Never threaten loss of creations, history, identity, relationships, earned value, portability, export, or account access to drive renewal, return, referral, review, share, or purchase—even when the underlying expiration/downgrade behavior is technically true. Communicate state neutrally and preserve the product's recovery/data-transition contract.
- Use product-native sharing, referrals, reputation, update education, and cross-promotion only when both sender and recipient receive real value. Raw shares, installs, messages, reviews, or opens are not success.
- Keep public review solicitation separate from private feedback and support. Platform review policy is adapter-specific; never sentiment-gate, star-steer, suppress unhappy users, reward ratings, or ask inside a platform that forbids it.
- Treat attribution as allocation evidence, not causality. Reserve holdouts/experiments where feasible, reconcile server/provider events, estimate incrementality, and use MMM or triangulation for channels that cannot be cleanly randomized.
- Optimize incremental retained contribution and user value with refund/fraud/support/trust countermetrics. Platform automated recommendations are proposals, not authority.
- Use constrained risk-adjusted decisions only after law, safety, consent, non-deception, claim truth, current platform authority, and bounded recoverability pass. An explicit prohibition is not a warning-priced experiment.
- Retrieve current platform/API/policy/asset/attribution facts at execution with publisher, canonical URL, version/scope, effective/retrieval/expiry dates, and digest. Never freeze volatile quotas, fields, products, windows, asset specs, review limits, or regional rules here.
Workflow
1. Establish product truth and marketing thesis
Capture product artifact IDs, user/job, category and alternatives, proof-backed claims, business model, release/channel availability, commercial constraints, audience modes, non-goals, and ruin boundaries. Define a falsifiable message hierarchy and acquisition/activation/retention relationship.
2. Build audience, consent, and lifecycle model
Create intent/lifecycle states, eligibility/exclusions, age/territory/sensitive-category treatment, first-party identity/consent, frequency/quiet-hour policy, suppression, deliverability, and recipient-value contract.
3. Build the selected channel and creative portfolio
Evaluate the role and handoff for owned, earned, paid, store, creator, community, partner, referral, lifecycle, reputation, cross-promotion, update education, SEO/AEO, and launch. Select channels from product truth, audience mechanism, recipient value, authority, and program target; build every selected adapter or specialist operating contract plus its localized brief and production-request supply now, route exact files through product-asset-production, and route official social account execution through social-media-operations-review, then choose activation from evidence and policy.
For a PC-store launch, preserve the separation between Store Listing narrative,
Product Asset Production files, Distribution build/store proof, Launch Admission
health, and Marketing's wishlist, creator, press, community and communication
program.
4. Define measurement and economics
Specify canonical exposure/cost/conversion/refund/fraud ledgers, identity/dedupe, attribution, experiments/holdouts, incrementality, contribution, uncertainty, saturation, payback, and trust/quality countermetrics. Assign action thresholds.
5. Define autonomous desired-state control
Create signed campaign/creative/spend manifests and separate proposer, creative, validator, promoter/reconciler, watchdog, causal evaluator, and recovery authorities. Use shadow checks, bounded canaries, maximum steps, hysteresis, cooldown, independent caps, live readback, and auto-pause.
6. Validate dangerous interactions
Test conflicting campaigns, cross-channel frequency, duplicate conversion, self-competition, expired claims, unavailable deep links, wrong locales, child/sensitive audiences, review/update/referral reward coupling, refund/chargeback, SDK consent drift, creator rights, invalid traffic, tracking loss, platform rejection, and cap divergence.
7. Close learning and recovery
Operate observe -> reconcile -> diagnose -> propose -> simulate -> validate -> canary -> readback -> causal_evaluate -> scale_or_pause -> learn. Every candidate and outcome is versioned; every pause/recovery preserves evidence and committed external state.
Source verification
Retrieve current advertising, measurement/attribution, privacy/consent,
child/age, sensitive-category, sponsorship/endorsement, creator/asset rights,
email/SMS/push, review/reputation, promotion/referral, platform automation,
spend/billing, territory, and channel API authority before activation. Explicit
prohibitions and missing authority are hard floors. Stale telemetry blocks
unbounded spend; it never justifies self-attested performance.
Hard gates
Reject or redesign an output that:
- cuts a selected channel, creative, measurement, localization, fraud, or automation capability because there are no users, uncertain demand, conventional staffing cost, or speculative ROI;
- enables every channel or spends broadly merely because adapters are built;
- invents product claims, prices, entitlement, offer, release, customer proof, endorsements, metrics, or availability;
- invents a digest, consumes an observed-state Product Program Manifest revision
that indexes Marketing, omits stable handoff IDs, resolves “latest” aliases,
back-references a same-revision consumer, or creates a circular graph;
- threatens loss of creations, history, identity, relationships, earned value, portability/export, or unrelated account access to drive renewal, return, referral, review, share, or purchase;
- optimizes clicks, installs, opens, followers, CPI, or platform ROAS without incremental retained-value economics and countermetrics;
- treats explicit platform prohibition, deceptive treatment, unlawful data, undisclosed sponsorship, fake review, spam, child targeting, or policy evasion as a priced grey-area canary;
- asks positive users for public reviews while diverting negative users privately, depicts five stars as steering, rewards reviews, or assumes Apple/Google review behavior on Steam;
- publishes scaled low-value search/social content, auto-replies indiscriminately, fabricates community engagement, or scrapes contacts;
- lets a platform recommendation, bidding model, creative agent, or proposer raise spend/caps, change claims, alter consent, or approve/promote itself;
- lacks independent daily/lifetime/product/channel/geo/campaign caps, stale-data failure, dedupe/reconciliation, and global pause;
- initializes tracking/export audiences without consent/eligibility, lets a dormant-never-activated channel create data/network/token/spend cost, or lets a paused/withdrawing adapter create new delivery/spend beyond reconciliation-only authority;
- claims causal lift from attribution alone or ignores holdouts, interference, novelty, seasonality, selection, fraud, refunds, saturation, and support;
- omits creative rights/provenance, accessibility, localization/cultural safety, AI/sponsorship disclosure, brand safety, inventory transparency, deep-link health, or release compatibility;
- relies on a routine human operator rather than the owning automated adapter or specialist to publish, bid, tune, reconcile, pause, refresh APIs, or recover incidents;
- claims production proof without exact candidate IDs, platform observed state, spend/cost reconciliation, live creative/landing readback, and downstream retained-value data.
Output contract
Produce one Marketing Automation Blueprint containing:
- draft artifact identity or sealed envelope and consumed product/claim/offer/release registry, with digests only for sealed inputs;
- objective hierarchy, thesis, audience/intent/lifecycle/consent model, exclusions, non-goals, and ruin boundaries;
- full channel sweep plus selected capability portfolio, with inclusion/exclusion rationale, role, adapter authority, scale/spend/failure envelope, construction/proof/activation state, dormant contract, and handoffs;
- positioning/message translation and proof-backed claim registry references;
- organic/content/SEO/AEO/social/community/creator/PR/partner/store/lifecycle/referral/cross-promotion/update/reputation integration plan, including typed Social Media Operating Model handoffs and the PC-store launch calendar and communication program when selected;
- paid channel adapter and immutable bid/budget/spend envelope matrix;
- localized creative brief and production-request graph with message/variant
lineage, provenance and rights requirements, accessibility, disclosures,
fatigue, acceptance gates, and handoffs to
product-asset-production;
- event/cost/identity/dedupe/attribution/experiment/incrementality/MMM/economics measurement plan;
- privacy/consent/age/territory/sensitive-category, fraud/IVT, brand safety, reputation, deliverability, complaint/refund/support guardrails;
- desired-state manifests, role/authority separation, canary/promotion/pause policy, reconciliation, incident and recovery model;
- specialist artifact handoffs for campaigns, referrals, listings, reviews/feedback, notifications, positioning, and distribution;
- dangerous-interaction tests, current authority records, validation roadmap, unresolved hypotheses, and next proofs.
Routing boundaries
marketing-automation-blueprint owns the complete multi-channel marketing operating system and control plane.
social-media-operations-review owns recurring account, publishing/readback, listening, reply/moderation routing, crisis, impersonation, rights, recovery, and shutdown operations only after Marketing selects the official social channels, audience and message.
market-research-synthesis owns source-bounded market/positioning evidence; marketing consumes and channel-translates it.
promotion-campaign-review owns one offer, event, update-reward, cross-promotion, or lifecycle campaign.
referral-loop-review owns qualification, attribution, pending grants, reversals, and referral fraud.
store-listing-optimization owns listing metadata/creative sequence/conversion; software-distribution-readiness owns submission/release evidence.
product-asset-production owns deterministic product capture, localization,
exact rendered media, rights/provenance, accessibility variants, and file QA.
review-solicitation-policy owns platform-specific authentic public review
eligibility and request state; product-feedback-learning-loop owns universal
private feedback, review ingestion, support/product routing and close-loop.
notification-strategy-review owns notification channel/copy/frequency semantics; promotion-campaign-review owns one dormant-user recovery campaign; market-research-synthesis owns source-bounded voice-of-customer evidence.
app-design-blueprint/game-design-blueprint own product-side share, cross-promotion, notification, review, and lifecycle surfaces.
Completion check
The blueprint is executable only when the full channel landscape has been
assessed, every selected channel and creative brief has a complete target and
stable production handoff, and channel operation can be disabled at zero cost,
activated through exact versioned state, reconciled to observed platform/cost
truth, bounded by spend/consent/claim/safety gates, measured for incrementality
and retained value, and automatically paused/recovered. It must remain
impossible for Marketing to invent product truth, render a competing product
media source, or approve its own authority.