Media Acquisition Virtuoso — Boss Agent and MOA Registration

Version: 1.0.0
Registered: August 5, 2026
Status: Code-side registration complete on the feature branch; production activation follows the normal Boss/MOA deployment pipeline.

What is registered

The shared acquisition skill defines five expertise overlays:

Profile	Runtime purpose
`media_acquisition_virtuoso`	Cross-channel strategy, unit economics, incrementality, and specialist orchestration
`google_ads_virtuoso`	Google Ads, GA4, Merchant Center/feed, Search, Shopping, PMax, bidding, and experiments
`meta_ads_virtuoso`	Meta Ads, Pixel+CAPI, Advantage+, audiences, creative systems, fatigue, and lift
`email_lifecycle_virtuoso`	Deliverability, consent/suppression, lifecycle flows, campaigns, and incremental revenue
`shopify_commerce_virtuoso`	Shopify catalog, inventory, orders, customers, discounts, returns, merchandising, CRO, and LTV

The profile names are skill overlays, not model IDs. Model selection remains governed by the central Virtuoso role registry.

Registration chain

skills/media-acquisition-virtuoso/SKILL.md
skills/media-acquisition-virtuoso/references/profile_registry.yaml
        │
        ├── .claude/skills/media-acquisition-virtuoso/**   (Claude Code parity)
        ├── src/shared/virtuoso_models/skills_data/media_acquisition_virtuoso/**
        ├── services/virtuoso-models-service/.../skills_data/media_acquisition_virtuoso/**
        │
        └── inline registration in the canonical Boss skillpack
             skills/miz-oki-platform-expert/references/boss_agent_skillpack.md
                    │
                    ├── Boss startup: skillpack_bootstrap.compose_skillpack_context()
                    ├── MOA experts: model_factory._moa_system_instruction()
                    ├── MCP: skills_get_section("10")
                    ├── src/shared/virtuoso_models/skills_data/...
                    └── services/virtuoso-models-service/.../skills_data/...

The five specialist entries are also registered in miz-oki-adk-agents/config/config.yaml under agents.moa_controller.specialists. Each points to the canonical skill source, uses a governed model_role, and is pinned to recommend_only.

Routing behavior

A single-channel request selects the matching channel specialist.
A request involving two or more channels, budget allocation, CAC/LTV, attribution conflict, or acquisition strategy selects media_acquisition_virtuoso first.
The cross-channel profile delegates to the relevant channel profiles.
Boss Agent orchestrates evidence collection and produces the approval-ready decision package.
MOA specialists evaluate independently and preserve material dissent; consensus cannot bypass a failed validation gate.

Governed operating boundary

All provider evidence follows:

provider API/webhook
  → service-marketing-connectors
  → service-canonical-ingestion
  → canonical event envelope
  → attribution / causal / policy / DCP layers

There is no direct connector-to-KG, connector-to-BigQuery, connector-to-Firestore, or connector-to-activation path.

Platform-reported ROAS and attribution are labeled as reported evidence. Cross-channel decisions use incremental revenue or contribution profit, confidence intervals, and iROAS where supported.

Spend, bids, budgets, audiences, catalogs/feeds, discounts, and lifecycle sends remain recommend-only until all required controls are present:

current evidence and data-quality checks;
registered holdout/experiment;
validation passport;
DCP authorization;
two-key human approval;
idempotency key;
pre-state capture;
rollback plan;
provenance/audit record.

Legacy quarantine

The skill explicitly quarantines these legacy acquisition-playbook tools for production decisions:

playbook_create_campaign
playbook_sync_audience
playbook_trigger_flow
playbook_query_metric

The legacy module contains direct platform paths and mock metric behavior. These tools may not be treated as production truth or used for production mutation until they are wrapped by current canonical-ingestion, no-mock evidence, DCP, approval, idempotency, and rollback controls.

Verification

Run from the repository root:

python scripts/acquisition_skills_sync.py --check --json
python scripts/skills_sync.py --check
pytest -q tests/skills/test_media_acquisition_virtuoso.py

Expected acquisition parity result:

Media Acquisition Virtuoso parity: OK (5 profiles; governance preserved)

The dedicated CI workflow .github/workflows/media-acquisition-skills-parity.yml runs these gates for every relevant change and on manual dispatch.

Production rollout

Merge through the normal reviewed PR path. The existing deploy router watches the shared skillpack/model-registry paths and routes the affected Boss/MOA deployment workflows. After deployment, verify:

# Boss skillpack parity
curl -s -X POST "$BOSS_URL/execute" \
  -H "Content-Type: application/json" \
  -d '{"tool":"skills_get_section","section_name":"10"}'

# Confirm the response contains all five profile IDs and the
# "Acquisition Virtuoso Registry v1.0" marker.

# MOA health/consensus metadata should report the same skillpack hash
# as the Boss runtime.

Do not claim production registration until the latest created and ready Cloud Run revisions match and runtime section retrieval shows the new registry.