marketinga.agent — marketing capabilities on the CEO + Seven Soldiers boardroom

The marketing Counsellor is not a parallel cabinet. It is the existing CEO + Seven Soldiers boardroom (daio/governance/boardroom.py:Boardroom) with each soldier carrying a marketing-specific skill. A campaign brief becomes a boardroom directive; each soldier votes; soldiers whose vote was approve AND who own a marketing skill execute that skill; per-soldier outputs assemble into the campaign envelope; the CEO signs the MarketingAttributionReceipt post-consensus, with the BoardroomSession id now an indexed on-chain field.

This is HBR's "layered system of specialized agents" thesis instantiated on the boardroom we already operate.

The cabinet IS the boardroom

                              CEO (brief composer + post-consensus signer)
                                            │
              ┌─────────┬─────────┬─────────┼─────────┬─────────┬─────────┐
            CPO       CTO       COO       CFO       CISO       CLO       CRO
          (1.0)     (1.0)     (1.0)     (1.0)    (1.2 veto) (0.8)   (1.2 veto)
            │         │         │         │         │         │         │
         content   experiment dist'n   reporting  voice      regs     spend
          (HBR1)   (HBR2)    (HBR3)    (HBR4)     gate     gate      gate

Soldier	Weight	Marketing skill	Skill module
CEO	n/a	Brief composition + consensus signer	`agents/marketing/skills/ceo.py`
CPO Product	1.0	HBR L1 — content drafting	`agents/marketing/skills/cpo.py`
CTO Technology	1.0	HBR L2 — experimentation	`agents/marketing/skills/cto.py`
COO Operations	1.0	HBR L3 — distribution	`agents/marketing/skills/coo.py`
CFO Finance	1.0	HBR L4 — reporting + treasury	`agents/marketing/skills/cfo.py`
CISO Security	1.2× veto	Identity gate + voice violation scan	`agents/marketing/skills/ciso.py`
CLO Legal	0.8	Regulatory + competitor neutrality	`agents/marketing/skills/clo.py`
CRO Risk	1.2× veto	Spend risk + hard-stop	`agents/marketing/skills/cro.py`

agents/marketing/skills/registry.py is the single source of truth for soldier_id → skill instance. It is tested directly in tests/test_marketing_skill_registry.py.

Brand code

Every cycle reads from data/brand_code/:

voice.md — references docs/MANIFESTO.md + docs/THESIS.md
positioning/pillars.md — six pillars; the sixth (code_as_dojo) is reserved for the human Convener
positioning/icp_segments.md — five concentric audience rings (E reserved for human)
forbidden_terms.json — token-shill ban list (regex compiled with re.IGNORECASE)
competitor_map.json — neutral-comparison rules per competitor
regulatory_constraints.md — what marketinga.agent must NOT promise
onboarding/{ceo,cpo,cto,coo,cfo,ciso,clo,cro}.marketing.md — Joseph Fuller HBR job descriptions, one per soldier role

Boardroom session lifecycle

1. CEO composes CampaignBrief + formats boardroom directive
Boardroom.convene(directive, members="ceo,cpo,cto,coo,cfo,ciso,clo,cro")
   ↓ each soldier votes (pure inference; no tools)
CISO veto check: if ciso_security.vote == "reject" → REFUSE
CRO veto check:  if cro_risk.vote == "reject"     → REFUSE
Overall outcome check: if session.outcome != "approved" → REFUSE
For each soldier with vote == "approve" AND a registered skill, in DISPATCH_ORDER:
     a. ciso_security  — identity + voice scan        (defense in depth)
     b. clo_legal      — regulatory + competitor
     c. cpo_product    — content draft (HBR L1)
     d. cto_technology — variants + holdouts (HBR L2)
     e. coo_operations — distribution (HBR L3)
     f. cfo_finance    — GEO probe + KPI snapshot (HBR L4) + treasury read
     g. cro_risk       — holdout integrity + spend hard-stop
   Each skill emits one marketing.soldier_skill_executed catalogue event.
   Each executing soldier gets ONE Tessera credential (per-soldier provenance).
CEO assembles CampaignEnvelope (with boardroom_session_id) and signs the
   
MarketingAttributionReceipt — emits the marketing.campaign_executed event.

CISO and CRO carry hard veto: a vote == "reject" from either short-circuits regardless of weighted score. CLO has no hard veto — its reject only blocks the campaign if the weighted score also falls below supermajority.

Three-receipt model

A campaign produces three on-chain receipt types — see MARKETING_RECEIPTS.md for the full explainer. Briefly:

Receipt	Layer	Per
`Tessera.sol`	identity	per soldier action
`X402Receipt.sol`	payment	per paid call
`MarketingAttributionReceipt.sol`	attribution	per campaign envelope (now keyed by `boardroomSessionId`)

Plus MarketingTreasury.sol for the 99% revenue → BANKON SATOSHI buyback/burn rule.

Backend surface

GET /marketing/status                       # 8-row soldier ↔ skill table
GET /marketing/campaigns                    # recent decisions from catalogue
GET /marketing/brand_code                   # current loaded brand-code (read-only)
GET /marketing/geo                          # last GEO probe rollup
GET /marketing/identity                     # on-chain identity binding status
GET /marketing/session/{boardroom_id}       # join BoardroomSession with the receipt + skill outputs

Every endpoint accepts ?h=true for plain-text rendering via mindx_backend_service.text_render.

Configuration

data/config/marketinga.toml — inotify-watched. Notable thresholds:

governance.spend_threshold_usd — Boardroom routing flag ($500)
risks.hard_stop_spend_usd — kill switch ($5,000)
reporting.geo_weekly_budget_usd — GEO probe spend cap ($20)
experimentation.holdout_rate — measurement holdout (10%)

On-chain identity binding

Phase 1 ships in dry-run mode: 8 ENS subnames (ceo.bankon.eth, cpo.bankon.eth, …, cro.bankon.eth) are unregistered until the operator runs:

python -m agents.marketing.onchain.bind_identity --execute

This binds:

Tessera credentials for each agent (CEO + 7 soldiers).
ENS subnames under bankon.eth.
AgentRegistry (ERC-8004) registrations with capability bitmaps.
iNFT_7857 mints — one per soldier seat.

The MarketingAttributionReceipt and MarketingTreasury contracts deploy via daio/contracts/marketing/script/DeployMarketing.s.sol — see daio/contracts/marketing/README.md for the operator command.

Phase 1 vs Phase 2

In Phase 1 (this drop):

8 soldier skills wired; orchestrator drives boardroom + dispatches skills end-to-end in dry-run.
Outbox-only distribution; no live publishes.
Mocked GEO probes possible (test path); production probes use the boardroom's existing inference cascade — never pinning a model.
Catalogue events emitted at every boundary (6 marketing.* event kinds).
Foundry tests at 50,000 fuzz runs green (incl. the new boardroomSessionId invariant).
Plain-text mode on every backend endpoint.

Operator-gated Phase 2:

Mainnet deploy of MarketingAttributionReceipt (Base, schema v2) + MarketingTreasury (Ethereum L1).
ENS subname registration ceremony for 8 soldier seats.
Per-soldier Tessera DID issuance.
iNFT_7857 mints.
Live MINDX_MARKETING_FARCASTER_LIVE, MINDX_MARKETING_X_LIVE flags.

Out of scope (separate plans):

Modifying Boardroom.convene itself — soldiers stay pure voters.
Senatus / Curia / Genius / Tabularium / Fides / SponsioPactum — separate constitutional-layer plan.
Kuzu / Qdrant / Meilisearch / NATS substrate — catalogue Phase 1+ plan.
90-day campaign execution — see MARKETING_PLAYBOOK.md.

Verification

# Python tests (59 passing as of this drop)
.mindx_env/bin/pytest -q --override-ini="addopts=" \
  tests/test_marketing_catalogue_events.py \
  tests/test_brand_code_loader.py \
  tests/test_geo_probe.py \
  tests/test_marketing_onchain_clients.py \
  tests/test_marketinga_agent.py \
  tests/test_marketing_soldier_skills.py \
  tests/test_marketing_skill_registry.py
Foundry CI gate (23 tests at 50k fuzz runs)
cd daio/contracts && FOUNDRY_PROFILE=marketing forge test --fuzz-runs 50000 -vvv
Live backend smoke
curl http://localhost:8000/marketing/status?h=true     # 8-row soldier mapping
curl http://localhost:8000/marketing/session/<id>?h=true

Referenced in this document

MANIFESTO MARKETING MARKETING_PLAYBOOK MARKETING_RECEIPTS README THESIS

All Documents Document Index The Book of mindX Improvement Journal API Reference