Marketing Council

Stop asking AI for marketing ideas. Make it defend a marketing decision.

29 Agent Skills. 24 specialist agents. 25 marketing figures. 46 applied theories. One connected decision process.

Marketing Council makes ChatGPT, Codex, Claude Code, and compatible AI agents diagnose the business problem, compare competing strategic views, check evidence and economics, challenge weak assumptions, and only then recommend tactics.

Version Agent Skills Specialist Agents License

The Neural Marketing Graph

Marketing Council v1.2 adds an explicit connection graph instead of leaving theory selection hidden inside prompts.

HOOKS -> SIGNALS -> SCHOOLS / THEORIES -> AGENTS -> SKILLS -> CHALLENGE -> DECISION

The graph currently connects:

  • 25 marketing figures and practitioners as source-linked decision cards
  • 19 schools of thought
  • 44 principles
  • 46 applied theories
  • 23 diagnostic signals
  • 24 specialist agents
  • 29 focused skills
  • 14 challenge hooks
  • 682 explicit connections

This means a mature category with weak differentiation can activate positioning and competitive-strategy specialists, while a conversion problem caused by friction can route toward behavioral diagnosis instead of automatically asking for stronger copy.

Try the deterministic router:

python scripts/neural_router.py \
  --signals category-mature,differentiation-weak,competitor-pressure-high \
  --json

The graph also stores explicit counterweights. A narrow-entry-audience principle can be challenged by penetration-growth logic; short-response accountability can be challenged by longer-horizon effectiveness; positioning focus can be challenged by broad mental-availability requirements.

Marketing schools in the graph

The figure cards are not personas. They are source pointers into different ways of making marketing decisions.

School Figures represented What it changes
Marketing management Philip Kotler, Kevin Lane Keller Market definition, STP, value and mix decisions
Customer-centered market definition Theodore Levitt, Peter Drucker What business and customer problem are actually being served
Positioning and category strategy Al Ries, Jack Trout, April Dunford Competitive alternatives, category frame, reason to prefer
Direct response and scientific advertising Claude Hopkins, David Ogilvy, Rosser Reeves Offer, proof, proposition, testing, response
Awareness and sophistication Eugene Schwartz Message depth, mechanism, proof burden, directness
Marketing science and brand growth Andrew Ehrenberg, Byron Sharp Penetration, mental availability, physical availability, buying patterns
Brand equity David Aaker, Kevin Lane Keller Associations, salience, perceived quality, brand architecture
Behavioral science and choice design Robert Cialdini, BJ Fogg, Richard Thaler, Rory Sutherland Friction, prompts, defaults, framing, uncertainty
Creative advertising Bill Bernbach, David Ogilvy, Rory Sutherland Translation from strategy into attention and memorable communication
Product narrative and demonstration Steve Jobs Focus, demonstration, product truth, launch story
Permission and remarkability Seth Godin Entry audience, relevance, voluntary attention
Competitive strategy Michael Porter Substitutes, rivalry, bargaining power, defensibility
Marketing effectiveness Les Binet, Peter Field Short and longer horizons, business effects, objective balance
Jobs to Be Done Clayton Christensen Buying circumstances, progress, switching triggers, alternatives

A named figure can never settle a debate. The graph carries the principle into the decision, then applies a counterweight when another school has a credible competing explanation.

2026 AI-mediated marketing layer

Version 1.3 adds a dated evidence layer for market changes that should not be confused with timeless marketing theory. The neural router can now activate dedicated paths for:

  • AI-mediated search, answer surfaces, and conversational advertising
  • Agentic commerce, machine-readable product truth, and commerce-feed readiness
  • Autonomous media decision rights, signal quality, and rollback governance
  • Incrementality, MMM, counterfactual design, and closed-loop bias
  • Creator commerce and creator-role-specific measurement
  • Commerce media across retailers, marketplaces, social commerce, and purchase-data environments
  • Synthetic creative hypothesis families and provenance checks

Current platform claims live under references/2026/ with dated source IDs. The Council must re-check those facts when they are load-bearing.

The problem

Most AI marketing sessions collapse into the same pattern:

"Try short-form video. Build awareness. Create valuable content. Test different creatives. Track KPIs."

That is not strategy. It is a list of familiar activities.

Marketing Council changes the sequence.

UNDERSTAND
  -> DIAGNOSE
  -> RESEARCH
  -> DISPATCH
  -> DEBATE
  -> CHALLENGE
  -> DECIDE
  -> PLAN
  -> EXECUTE
  -> MEASURE
  -> LEARN

If the evidence says the requested tactic is wrong, the Council should reject it. If the economics do not support the campaign, it should say so. If two marketing schools point in different directions, it should preserve the disagreement until the deciding factors are clear.

Install in 30 seconds

Any agent supported by Skills CLI

Install every Marketing Council skill to every supported agent:

npx skills add imMamdouhaboammar/marketing-council-pack --all -y

See what the repository contains before installing:

npx skills add imMamdouhaboammar/marketing-council-pack --list

Install all skills globally for Codex only:

npx skills add imMamdouhaboammar/marketing-council-pack --skill '*' -g -a codex -y

Install all skills globally for Claude Code only:

npx skills add imMamdouhaboammar/marketing-council-pack --skill '*' -g -a claude-code -y

Install only the main Council skill:

npx skills add imMamdouhaboammar/marketing-council-pack --skill marketing-council -g -y

Run the main skill without installing it:

npx skills use imMamdouhaboammar/marketing-council-pack --skill marketing-council --agent codex

Update installed global skills:

npx skills update -g -y

npx skills installs Agent Skills. For Claude's 18 native specialist subagents, use the Claude marketplace installation below.

ChatGPT + Codex plugin

Marketing Council includes a native .codex-plugin/plugin.json, OpenAI metadata for all 29 skills, repository marketplace metadata, brand assets, and the 24 specialist role files used by the Council.

Add the repository marketplace:

codex plugin marketplace add imMamdouhaboammar/marketing-council-pack

Inspect or refresh it:

codex plugin marketplace list
codex plugin marketplace upgrade marketing-council

In the ChatGPT desktop app, open the Plugins Directory, select the Marketing Council marketplace, and install the plugin.

Build a standalone OpenAI plugin ZIP:

python scripts/build_host_packages.py

Output:

dist/release/marketing-council-openai-plugin-v1.3.0.zip

The OpenAI package is skills-only. It does not invent an MCP server or executable lifecycle hook just to make the manifest look more complex.

Claude Code marketplace

The Claude plugin exposes both sides of Marketing Council:

  • all 29 skills under skills/
  • all 18 role definitions as native Claude subagents under agents/

Inside Claude Code:

/plugin marketplace add imMamdouhaboammar/marketing-council-pack
/plugin install marketing-council@marketing-council

Validate a local checkout:

claude plugin validate . --strict

Build a standalone local marketplace ZIP:

python scripts/build_host_packages.py

Output:

dist/release/marketing-council-claude-marketplace-v1.3.0.zip

That archive is self-contained. Its marketplace points to a local ./plugins/marketing-council copy, so the plugin, skills, agents, references, and scripts travel together.

What happens when you ask a marketing question

Say you ask:

"Conversion is weak. I want to cut price by 20% and go hard on TikTok. Build the plan."

A generic assistant can start writing TikTok ideas immediately.

Marketing Council should first work out whether the real problem is demand, traffic quality, offer clarity, pricing, checkout friction, product fit, or retention. It can then route the same evidence brief to the relevant roles, for example:

market-architect
commercial-strategist
channel-strategist
response-strategist
marketing-skeptic

The Council then has to answer questions such as:

  • Is price actually suppressing conversion?
  • What happens to contribution margin and CAC payback after the cut?
  • Is TikTok where this buying situation happens, or merely the requested channel?
  • What evidence supports the audience claim?
  • What would make us abandon the recommendation?
  • What are we explicitly choosing not to do?

The result is a decision with assumptions, risks, thresholds, and next actions, not a longer brainstorm.

Built from marketing schools, not celebrity role-play

Marketing Council does not assign celebrity identities to the AI model. Philip Kotler, Seth Godin, David Ogilvy, Claude Hopkins, Eugene Schwartz, Al Ries, Jack Trout, Robert Cialdini, Byron Sharp, Les Binet, Peter Field, Steve Jobs, and other named thinkers appear only as sources or reference points for decision principles.

Instead, the repository turns useful ideas associated with established marketing schools into decision cards with four practical fields:

Principle
When it applies
When it should not be over-applied
What evidence or competing principle should challenge it

Examples live in references/canon/.

The point is not asking for a famous marketer's imagined response. The useful question is:

Which principle fits this evidence, under these market conditions and commercial constraints, and what would falsify the recommendation?

This is an independent project and is not affiliated with or endorsed by the authors, estates, publishers, or companies referenced in the principle library.

The Council

Market and customer

Agent Job
market-architect Market definition, segmentation, category structure, demand, route to market
audience-strategist Buying situations, jobs, triggers, anxieties, alternatives, customer language
positioning-strategist Category frame, alternatives, differentiation, memory
product-marketing-director Product truth, focus, demonstration, launch narrative

Persuasion and behavior

Agent Job
response-strategist Offer, proof, objections, CTA, measurable response
awareness-strategist Awareness, sophistication, message depth, proof requirements
behavior-strategist Friction, choice architecture, risk, social proof, decision cues

Growth and commercial reality

Agent Job
brand-growth-strategist Reach, penetration, availability, distinctive assets, time horizon
commercial-strategist Price, margin, CAC, LTV, payback, retention, sales capacity
channel-strategist Media, search, creators, partnerships, distribution and channel fit

Governance

Agent Job
marketing-skeptic Finds weak evidence, hidden assumptions, channel bias, fake certainty
council-director Selects roles, preserves disagreements, resolves the final decision

14 installable skills

The repository is deliberately modular. A narrow task should not load an entire strategy engagement.

marketing-council
market-diagnosis
customer-research
positioning-strategy
offer-strategy
pricing-strategy
go-to-market
campaign-strategy
media-strategy
content-strategy
conversion-strategy
retention-strategy
marketing-experimentation
competitive-intelligence

The main marketing-council skill routes broad or conflicted work. The other 13 skills handle focused jobs with less context.

Eight challenge gates

Before a significant recommendation is accepted, the pack can apply:

  1. strategy-before-tactics
  2. evidence-gate
  3. freshness-check
  4. commercial-reality-check
  5. customer-language-check
  6. anti-generic-marketing
  7. pre-mortem
  8. post-strategy-red-team

These are reasoning gates stored as marketing guidance. They are not hidden shell commands or auto-running lifecycle hooks.

Evidence has a status

Load-bearing statements can be labeled as:

FACT
EVIDENCE
INFERENCE
ASSUMPTION
HYPOTHESIS
UNKNOWN

That small constraint matters. "Customers value authenticity" cannot quietly become customer research when nobody actually observed it.

Strategy means choosing

A full Council output can cover:

  1. Situation
  2. Decision to make
  3. Evidence map
  4. Diagnosis
  5. Audience and buying situation
  6. Positioning and category frame
  7. Value proposition and offer
  8. Primary strategic choice
  9. What we will not do
  10. Channel and distribution strategy
  11. Message architecture
  12. Prioritized tactics
  13. Experiments
  14. Measurement and thresholds
  15. Risks and assumptions
  16. Next decisions

The ninth item is intentional. A strategy that refuses to exclude anything is usually a backlog.

Tactics need a mechanism

Every recommended tactic is expected to define:

objective
customer / audience
evidence or insight
mechanism
message
channel
desired action
expected effect
cost or effort
primary risk
measurement
success threshold
failure threshold
next action if it works
next action if it does not

If a tactic cannot explain why it should work or how failure will be detected, it is not ready.

Deterministic marketing utilities

Some decisions should use arithmetic instead of prose. The pack includes dependency-free Python utilities for:

  • unit economics
  • funnel rates
  • experiment sample-size planning
  • tactic ranking
  • strategy linting
  • pack and distribution validation

Examples:

python scripts/unit_economics.py \
  --revenue-per-order 100 \
  --cogs-per-order 30 \
  --variable-costs-per-order 10 \
  --cac 25 \
  --expected-orders-per-customer 2

python scripts/funnel_math.py visits=1000 leads=100 customers=20
python scripts/experiment_math.py plan --baseline-rate 0.10 --mde 0.02
python scripts/strategy_linter.py path/to/strategy.md

Tool contracts, not hard-coded vendors

Marketing Council describes capabilities such as:

web.search
web.fetch
files.search
files.read
analytics.query
ads.query
crm.query
search-console.query
spreadsheet.calculate

A host can bind those capabilities to whatever tools it actually has. If a capability is missing, the skill should expose the evidence gap rather than pretend the data was checked.

See tools/capabilities.yml.

Evals are part of the product

The repository includes 8 business cases and 6 adversarial cases.

They test whether the agent:

  • diagnoses before prescribing
  • changes its recommendation when the business conditions change
  • separates evidence from assumptions
  • lets economics affect the decision
  • distinguishes strategy from tactics
  • rejects requested-channel bias
  • refuses invented customer insight and fake scarcity
  • does not treat attribution as causation
  • exposes uncertainty

See evals/.

Build releases

Build every host package plus the source archive:

python scripts/build_host_packages.py

Generated files:

marketing-council-openai-plugin-v1.3.0.zip
marketing-council-claude-marketplace-v1.3.0.zip
marketing-council-skill-v1.3.0.zip
marketing-council-pack-v1.3.0.zip
marketing-council-v1.3.0-SHA256SUMS.txt

Build only the single self-contained Agent Skill directory:

python scripts/build_dist.py

Validate

python -m unittest discover -s tests -v
python scripts/validate_pack.py
python scripts/validate_distribution.py --json

For an extracted OpenAI plugin package:

python scripts/validate_openai_plugin.py path/to/extracted/plugin --json

When Claude Code is installed locally:

claude plugin validate . --strict

Repository map

marketing-council-pack/
├── .agents/plugins/          # ChatGPT/Codex repo marketplace
├── .codex-plugin/            # ChatGPT/Codex plugin manifest
├── .claude-plugin/           # Claude plugin + marketplace manifests
├── assets/                   # Plugin identity
├── skills/                   # 29 Agent Skills
├── agents/                   # 24 specialist role definitions
├── hooks/                    # Marketing challenge gates, not lifecycle hooks
├── workflows/                # Full strategy, launch, campaign, audit, debate
├── references/               # Figures, schools, principles, theories, canon, frameworks
├── neural/                   # Knowledge graph, signals, routing guide
├── tools/                    # Host-neutral tool capability contracts
├── scripts/                  # Math, linting, builders, validators
├── evals/                    # Core and adversarial scenarios
├── examples/                 # Example strategy and council debate
├── adapters/                 # Host-specific usage notes
└── tests/                    # Structural and behavioral checks

Good prompts to start with

Diagnose this marketing problem before recommending tactics: [context]

Run a council debate on this strategy. Show where the specialists disagree, what decides the disagreement, and what evidence would reverse the final recommendation: [strategy]

Red-team this marketing plan. Find unsupported assumptions, commercial risks, channel bias, and weak measurement. Then rebuild only the parts that fail: [plan]

Build a go-to-market strategy for [product] in [market]. Separate facts, evidence, assumptions, hypotheses, and unknowns. Research current facts before making them load-bearing.

Security and privacy

Marketing Council ships without credentials, tracking code, an MCP server, or executable lifecycle hooks. Host tools and connected data remain subject to the host's own permissions and user authorization.

See SECURITY.md.

License

MIT. See LICENSE.