WorkOS: AI Skill System Architecture

The CEO skill handles decisions that can't be delegated: capital allocation, org design, the calls that keep you up at night. When you say /ceo, it reads data from every other skill because the CEO needs the full picture.

The 7-Step Lifecycle

1️⃣

Invoke: /ceo

Type the command from any channel

▶

You say /ceo "should we raise now or wait?". Gateway loads skills/ceo/SKILL.md — 421 lines of founder-CEO strategy, frameworks, and pushback protocols. The AI loads a wartime counselor persona that won't let you avoid the hard calls.

↓

2️⃣

Load the Full Picture

CEO reads ALL skill data

▶

📂 CEO's Own Data

strategy.json

Vision, current priorities, big bets, existential risks. Wartime or peacetime mode.

hard_conversations.json

Conversations you're avoiding. Who, why, and the cost of waiting.

decisions/

Major CEO decisions with context and outcomes.

board/

Board prep, meeting notes, investor relationships.

→

🔗 Reads from EVERYONE

CFO → data/cfo/

Runway, burn rate, forecast. "Are we going to run out of money?"

CMO → data/gtm/

Pipeline, win rate, GTM health. "Is the growth engine working?"

CPO → data/product/

PMF status, roadmap, retention. "Do customers love this?"

CTO → data/engineering/

Tech health, infra costs, team capacity. "Can we ship fast enough?"

👑 Why this matters

Other skills read 1-2 peers. The CEO reads all four. That's why it can catch misalignment between what the CFO says about runway and what the CMO is spending on pipeline.

↓

3️⃣

Wartime or Peacetime?

First assessment: what mode is the company in?

▶

⚔️ Wartime

Existential threat. Pivoting. Running out of money. Cash is tight, market shifted, competitor raised $50M.

CEO behavior: Direct, decide, cut. Speed over consensus. No peacetime luxuries.

🕊️ Peacetime

Growing. PMF achieved. Runway secure. Things are working and the challenge is scaling what's working.

CEO behavior: Delegate, develop, expand. Invest in culture and systems.

⚠️ Reality Check

Most startups are in wartime. Don't run a peacetime playbook in wartime. The skill checks strategy.json for the current mode and adjusts all advice accordingly.

↓

4️⃣

Apply CEO Frameworks

6 frameworks for the decisions only a CEO can make

▶

⚔️ Wartime vs Peacetime

Assess the mode. Different situations demand different CEO behavior.

📅 Calendar Audit

How should a CEO spend time? Strategy 20%, People 30%, External 25%, Ops 15%, Personal 10%.

🎯 Decision Rights

Type 1 (irreversible, CEO decides), Type 2 (reversible, delegate), Type 3 (don't even review).

💬 Hard Conversations

Track conversations you're avoiding. If it's been >2 weeks on the list, schedule it today.

🏛️ Board Management

No surprises ever. Share bad news with context AND a plan. Build relationships outside meetings.

💎 Capital Allocation

What are we buying? What are we NOT spending on? Success criteria. Kill criteria.

↓

5️⃣

The Sparring Protocol

Hard truths the CEO needs to hear

▶

Example Pushbacks:

"That's a decision you can delegate. What's the decision only YOU can make?"

"You're solving for this quarter. What does this look like in three years?"

"You're avoiding the hard conversation. Who do you need to talk to?"

"You know the answer. You're looking for permission. You don't need it."

The CEO skill challenges whether you should even be doing the work. If it's not a Type 1 decision, it pushes back: "Delegate this. Your job is direction, team, and capital. Everything else is someone else's job."

↓

6️⃣

Dual Output: Text + Data

CEO Sync document + strategy files

▶

💬 CEO Sync (for you)

1. State of the Company — honest wartime/peacetime assessment
2. The One Thing — single most important focus
3. Hard Truth — what you're probably avoiding
4. C-Suite Health — 🟢/🟡/🔴 per function (CMO, CFO, CPO, CTO)
5. Decisions Only You Can Make — pending Type 1 decisions
6. This Week — 2-3 specific CEO actions

📊 Strategy Files (for machines)

strategy.json — vision, priorities, mode, big bets, risks
hard_conversations.json — tracked conversations with deadlines
decisions/ — major decisions with context
sync_history.json — every CEO sync appended

These files feed the dashboard and future CEO sessions.

↓

7️⃣

Delegate to the C-Suite

CEO orchestrates. Everyone else executes.

▶

The CEO skill rarely does the work itself. It identifies what needs to happen and routes it:

/cfo

"We need a fundraise plan." → delegates to CFO + /fundraise-prep

/cmo

"GTM strategy needs adjustment." → delegates to CMO

/cpo

"Product isn't shipping fast enough." → delegates to CPO + CTO

/leadership-sync

"Get me a cross-functional view." → synthesizes all C-suite perspectives

The CFO skill turns the AI into "CJ" — a conversational, dad-joke-friendly finance leader who's rigorous underneath the banter. When You say /cfo, CJ loads financial frameworks, pulls data from across the company, and pushes back hard on bad assumptions. Every conversation produces both human advice and machine-readable JSON.

The 7-Step Lifecycle

1️⃣

Invoke: /cfo

Type a command in any channel

▶

You say /cfo "let's talk burn rate". Gateway loads skills/cfo/SKILL.md — 1,100 lines of finance expertise. The AI becomes CJ with all frameworks, benchmarks, and data paths loaded. Any channel works: Telegram, Discord, Claude Code terminal.

↓

2️⃣

Load Financial State

Read own data + pull from other skills

▶

📂 CFO's Own Data

assumptions.json

Revenue model: SaaS fees, FX take rate, yield on AUM. Scenario parameters. Valuation targets.

sync_history.json

Every past conversation with metrics. "What did we discuss last time?"

latest_forecast.json

Current 3-scenario financial model. The dashboard reads this file.

forecasts/forecast_*.json

Historical snapshots. How projections evolved over time.

→

🔗 Cross-Skill Reads

CMO → data/gtm/

Pipeline value, win rate, CAC by channel, marketing spend.

CPO → data/product/

PMF stage, AI cost per user, NPS, retention.

CTO → data/engineering/

Infra spend, headcount, open roles, tech debt.

💡 Why Cross-Skill Reads Matter

The CFO doesn't work in a vacuum. When you ask "what's our burn multiple?", the answer includes marketing spend from CMO, infra costs from CTO, and product health from CPO.

↓

3️⃣

First Run vs. Returning

Bootstrap if new, compare if returning

▶

🆕 First Run

No history found. CJ asks for baseline numbers: cash position, burn, MRR, client count, GTV, AUM, CAC, LTV, NRR, gross margin, AI inference spend. Creates data/cfo/ directory.

🔄 Returning

Reads last sync, compares metrics, calculates efficiency, recalculates 3-scenario forecast, challenges anything off, writes updated files, appends to history.

↓

4️⃣

Apply Financial Frameworks

9 embedded frameworks

▶

🔥 Burn Multiple

Net Burn ÷ Net New ARR. <1x amazing. >3x dangerous.

📊 Five Pillars

Growth + Retention + Margin + Sales Efficiency + Profitability.

📐 Rule of X

(Growth × 2-3x) + FCF Margin.

🎯 Sales Efficiency

Magic Number, CAC Payback, LTV/CAC.

📉 Tri-Scenario

Low / Medium / High on every forecast.

🔬 Unit Economics

Per-customer P&L after fully-loaded costs.

🤖 AI Economics

Cost per inference, AI margin impact.

🎭 Investor Personas

5 archetypes. Different pitch for each.

🗓️ Fundraise Timeline

4-phase roadmap with stage gates.

↓

5️⃣

The Sparring Protocol

CJ challenges every metric

▶

Example Pushbacks:

"Your burn multiple is 3x. That means you're spending $3 to generate $1 of ARR. That's not a growth story, that's a cash bonfire."

"You're showing me GMV, but I want to see gross margin. What's the actual unit economics on each transaction?"

"Rule of 40 looks fine, but Rule of X? You're underweighting growth."

The persona is "CJ" — conversational, uses dad jokes, irreverent about "boring" finance. Rigorous underneath the banter.

↓

6️⃣

Dual Output: Text + Data

Every interaction produces two things

▶

💬 For Humans

1. VC Reality Check — honest progress assessment
2. Numbers That Matter — metrics with framework analysis
3. Highest Leverage Action — THE one thing
4. Hard Questions — what VCs will ask

📊 For Machines

latest_forecast.json — 3-scenario model
forecasts/forecast_*.json — historical snapshot
sync_history.json — conversation data appended

Files power dashboards and future conversations.

↓

7️⃣

Delegate to Execution Skills

CFO strategizes. Child skills execute.

▶

/finance-forecast

Detailed scenario modeling. Revenue projections. Monthly granularity.

/cap-table

Equity tracking. Dilution analysis. Option pool modeling.

/board-deck

Quarterly board presentations. Metrics + narrative + asks.

/fundraise-prep

Data room. VC Q&A. Due diligence checklists.

The CMO skill is a composite GTM leader that blends the sharpest modern go-to-market thinking. Signal-obsessed, AI-native, and allergic to bloat. When You say /cmo, the AI becomes a strategic marketing partner who believes most 2023 playbooks are broken and that a founder with AI agents can outperform a 10-person GTM team.

The 7-Step Lifecycle

1️⃣

Invoke: /cmo

Ask a GTM question

▶

You say /cmo "our pipeline is thin, what do we do?". Gateway loads 630 lines of GTM strategy. The AI becomes a composite of the best modern GTM thinkers — signal-based selling, AI-native ops, positioning expertise, and bottom-up growth.

↓

2️⃣

Load GTM State

Marketing data + financial constraints + product roadmap

▶

📂 CMO's Own Data

project_context.json

Product, ICP segments, GTM model, stage, value props.

positioning.json

Dunford 5-component positioning framework.

gtm_scorecard.json

Pipeline metrics, content performance, customer health.

sync_history.json

Every past CMO conversation and strategic assessment.

→

🔗 Cross-Skill Reads

CFO → data/cfo/

Budget constraints, runway, revenue targets. GTM must align with financial reality.

CPO → data/product/

Product roadmap, upcoming launches. Content and positioning must match what's shipping.

↓

3️⃣

First Run vs. Returning

Discovery flow vs. strategic sync

▶

🆕 First Run

Runs full discovery: What's the product? Who's the buyer? How are you getting customers today? What channels have you tried? What tools are in the stack? What signals are you tracking?

🔄 Returning

Reads last sync, assesses GTM maturity stage (Explorer/Builder/Scaler), checks positioning clarity, identifies the one highest-leverage GTM action.

↓

4️⃣

Apply GTM Frameworks

10 frameworks for AI-era go-to-market

▶

📈 GTM Maturity

Explorer → Builder → Scaler. Stage-appropriate advice only.

🎯 Positioning (Dunford)

5 components: alternatives, capabilities, value, customers, category.

📦 Product Type

Pick ONE: Vertical, New Way, Buy vs Build, or 10x Better.

⚠️ Risk-Based Messaging

80% buy to avoid pain. Lead with risk, not aspiration.

⬆️ Bottom-Up GTM

Build user love first. Bottoms-up ocean feeds the top-down river.

📡 Signal-Based Selling

Website visits, intent data, engagement signals over spray-and-pray.

🤖 AI Growth Playbook

60-70% of traditional tactics are dead. Re-find PMF every 3 months.

🛠️ AI GTM Stack

Replace 10 SDRs with 1 person + AI agents.

💰 PLG Pricing

MOAT + DEEP frameworks. 12% freemium beats 8% free trial.

📊 Channel Priority 2026

P0: LinkedIn organic + signals + referrals. P2: paid ads (last).

↓

5️⃣

The Sparring Protocol

Anti-bloat, signal-obsessed pushback

▶

Example Pushbacks:

"That's a 2023 playbook. Here's what's actually working now..."

"Companies scale to $5M ARR on LinkedIn and signals alone. Have you tried that before hiring SDRs?"

"You're optimizing when you should be innovating. Ship a new feature instead of polishing old funnels."

"What are you positioning against? If you can't name the alternative, your positioning is weak."

↓

6️⃣

Dual Output: Text + Data

Strategic assessment + GTM scorecard

▶

💬 Strategic Assessment

1. Situation Read — where the business is in GTM journey
2. Positioning Check — is positioning clear? If not, priority #1
3. Top GTM Priority — THE one thing
4. Challenge — push back on an assumption
5. Next Moves — 2-3 executable steps

📊 GTM Scorecard JSON

gtm_scorecard.json — pipeline, content, customer, efficiency metrics
scorecards/scorecard_*.json — historical snapshots
positioning.json — Dunford 5-component framework
sync_history.json — every CMO sync appended

↓

7️⃣

Delegate to GTM Pipeline

11 execution skills + 6 orchestrated workflows

▶

The CMO has the largest child skill network — 11 execution skills covering the full GTM pipeline:

/gtm-icp

Define ICP segments and messaging

/gtm-prospecting

Build enriched prospect lists

/gtm-content

Generate segment-targeted content

/gtm-outbound

Execute outreach sequences

/gtm-lead-capture

Score and qualify leads

/gtm-deal-intel

Analyze deals, extract competitive intel

/gtm-onboarding

Post-close customer activation

/gtm-lifecycle

Expansion and retention playbooks

/gtm-analytics

Funnel diagnostics and attribution

/gtm-monetization

Packaging and pricing strategy

/gtm-infra

GTM tech stack and automation

/advisor-outreach

Network-based intro harvesting

🔄 Orchestrated Workflows

The CMO can run coordinated multi-skill workflows: Strategy Mode (ICP → Monetization), Acquisition Mode (Prospecting → Content → Outbound → Lead Capture), Retention Mode (Onboarding → Lifecycle), and Full Funnel Review (Analytics across all stages).

The CPO skill is a strategic product leader obsessed with outcomes over outputs. When You say /cpo, the AI becomes customer-obsessed, anti-feature-factory, and model-aware. It blends timeless product wisdom (Cagan, Gibson Biddle, Doshi) with AI-native thinking (Model Maximalism, the Bottleneck Shift, the Ownership Principle).

The 7-Step Lifecycle

1️⃣

Invoke: /cpo

Ask a product question

▶

You say /cpo "should we build the sweep feature or the FX dashboard first?". Gateway loads 779 lines of product strategy. The AI becomes a product leader who thinks in outcomes, not features, and won't let you ship something nobody asked for.

↓

2️⃣

Load Product State

Product data + market context + financial constraints

▶

📂 CPO's Own Data

strategy.json

Product vision, PMF status, unfair advantage, biggest risks.

roadmap.json

Current priorities, upcoming features, RICE scores.

competitive_analysis.json

Direct competitors, indirect alternatives, positioning map.

pmf_tracker.json

Sean Ellis test results, retention curves, NPS trends.

→

🔗 Cross-Skill Reads

CMO → data/gtm/

ICP profiles, market positioning. What customers are saying and buying.

CFO → data/cfo/

Business model constraints, revenue targets, unit economics.

↓

3️⃣

First Run vs. Returning

Product discovery vs. strategy sync

▶

🆕 First Run

Full product discovery: What does it do? Who's it for? What pain does it solve? PMF status? What's working, what's not? AI core or feature? What becomes possible if models get 10x better?

🔄 Returning

Assesses PMF stage, checks DHM scores on current priorities, evaluates whether work is Leverage/Neutral/Overhead, challenges roadmap decisions.

↓

4️⃣

Apply Product Frameworks

13 frameworks — traditional + AI-native

▶

Traditional Product Wisdom:

📊 PMF Assessment

Pre-PMF → Emerging → Strong → Expanding. Don't scale what isn't working.

💎 DHM Framework

Delight × Hard-to-copy × Margin-enhancing. Score every initiative.

⚡ LNO Framework

Leverage (10%), Neutral (60%), Overhead (30%). Protect Leverage time.

📐 RICE + Strategy

Reach × Impact × Confidence / Effort. Plus vision alignment.

🔍 Discovery (Cagan)

4 risks: Value, Usability, Feasibility, Viability. Test riskiest first.

🗺️ Competitive Positioning

Direct, indirect, status quo. Why do deals stall?

AI-Native Product Thinking:

🚀 Model Maximalism

Build for emerging capabilities, not current limits. Today's models are the worst you'll ever use.

📏 Evals as Strategy

Product can only improve on what you measure. Build evals before features.

🔄 Bottleneck Shift

AI writes 90%+ of code. Decision-making is the new constraint.

🥚 Ownership Principle

Don't fully automate. Let users "add the egg." Small effort creates ownership.

📉 Value > Engagement

2-message success > 200-message struggle. Measure outcomes, not activity.

🏗️ Full Stack Builder

One person, idea to launch. Collapse silos. AI as force multiplier.

↓

5️⃣

The Sparring Protocol

Outcomes over outputs, teeth in opinions

▶

Example Pushbacks:

"That's a feature, not a strategy. What problem are we solving and for whom?"

"Is a chatbot actually the right interface, or are we defaulting to it because it's trendy?"

"That opinion has no teeth. What are you willing to sacrifice for this? What's the real tradeoff?"

"You say this is a priority, but where are the resources? Resource allocation reveals truth."

↓

6️⃣

Dual Output: Text + Data

Product assessment + strategy files

▶

💬 Product Assessment

1. PMF Status — which stage, evidence
2. Product Health — DHM scores on priorities
3. Top Product Priority — the one Leverage item
4. Challenge — what assumptions need testing
5. Next Moves — executable steps

📊 Product Data JSON

strategy.json — vision, PMF stage, unfair advantage
roadmap.json — prioritized features with RICE
competitive_analysis.json — landscape map
pmf_tracker.json — retention, NPS, Sean Ellis scores

↓

7️⃣

Delegate to Product Skills

Discovery before delivery

▶

/product-discovery

Validate assumptions through research, competitive analysis, build vs buy, feasibility.

/pm

PRD writing and feature specification with lean/MVP mindset for CTO-ready handoff.

The CTO skill is a pragmatic technical leader who believes in radical simplicity. When You say /cto, the AI becomes an engineering strategist who'd rather ship a monolith than debate microservices, who thinks boring technology wins, and who'll call out resume-driven decisions on sight.

The 7-Step Lifecycle

1️⃣

Invoke: /cto

Ask a technical question

▶

You say /cto "should we move to Kubernetes?". Gateway loads 476 lines of technical leadership. The AI becomes a CTO who designs for failure, champions simplicity, and won't let you over-engineer for a scale you don't have.

↓

2️⃣

Load Engineering State

Tech stack + budget + product requirements

▶

📂 CTO's Own Data

tech_stack.json

Frontend, backend, database, infra, CI/CD, observability. Maturity stage.

engineering_scorecard.json

DORA metrics: deploy frequency, lead time, change failure rate, MTTR.

tech_debt.json

Classified by quadrant: reckless/prudent × deliberate/inadvertent.

infra_costs.json

Cloud spend, cost per customer, compute efficiency, top cost drivers.

→

🔗 Cross-Skill Reads

CFO → data/cfo/

Budget constraints, runway. "Can we afford this infra investment?"

CPO → data/product/

Roadmap, upcoming features. "What do we need to build next?"

↓

3️⃣

First Run vs. Returning

Technical landscape mapping vs. health check

▶

🆕 First Run

Maps the full technical landscape: stack, team, deployment frequency, on-call rotation, pain points, infra budget, compliance requirements. Creates data/engineering/ directory.

🔄 Returning

Assesses maturity stage (Survival/Foundation/Scale/Optimize), reviews DORA metrics, checks tech debt trends, evaluates infra costs, identifies the one highest-leverage technical action.

↓

4️⃣

Apply Technical Frameworks

6 frameworks for pragmatic engineering

▶

🏗️ Technical Maturity

Survival → Foundation → Scale → Optimize. Stage-appropriate advice only.

📋 ADR Framework

Architecture Decision Records. Context, options, decision, consequences, review date.

🧱 Tech Debt Quadrant

Reckless/Prudent × Deliberate/Inadvertent. Classify to prioritize.

🛒 Build vs Buy

Core differentiator? Build. Everything else? Buy until proven otherwise.

💵 Infra Cost Model

Infra as % of revenue, cost per customer, compute efficiency.

🔒 Security Baseline

Minimum viable security per stage. From HTTPS to SOC 2 Type II.

↓

5️⃣

The Sparring Protocol

Radical simplicity, anti-complexity

▶

Example Pushbacks:

"That's over-engineered for your stage. What's the simplest thing that could work?"

"Before we add another service, show me the load that justifies it."

"This introduces operational complexity. Who's going to be on-call for it at 3am?"

"That's a resume-driven decision, not a business-driven one."

↓

6️⃣

Dual Output: Text + Data

Technical assessment + engineering scorecard

▶

💬 Technical Assessment

1. Situation Read — engineering health, maturity stage
2. Top Technical Priority — the one thing
3. Trade-off Acknowledgment — what you're NOT doing and why
4. Next Moves — 2-3 executable technical actions

📊 Engineering Scorecard JSON

engineering_scorecard.json — DORA metrics, debt, infra, security, team
tech_stack.json — architecture state
tech_debt.json — classified and prioritized
adrs/ — decision records
incidents/ — post-mortems

↓

7️⃣

Delegate to Engineering Skills

Specific technical execution

▶

/tech-debt

Track, prioritize, and plan debt paydown with impact-based classification.

/architecture-decision

Generate and review Architecture Decision Records.

/infra-cost

Analyze and optimize cloud infrastructure costs.