Stop Writing Prompts. Start Running Revenue: ai-marketing-skills

Your marketing team is bleeding money on AI theater.

You've bought the ChatGPT Plus subscriptions. You've built the "prompt libraries." You've sat through the webinars promising 10x productivity. And yet—your experiments still die in spreadsheets. Your content still ships with "AI slop" that tanks engagement. Your sales pipeline still depends on interns manually copying lead lists into Salesforce.

Here's the brutal truth nobody selling AI courses wants to admit: Prompts don't run businesses. Workflows do.

What if you could clone a complete marketing operating system—battle-tested on real pipelines generating millions in revenue—and deploy it with a single git clone? No more vibe-based marketing. No more "hope this prompt works." Just executable scripts, statistical rigor, and automation pipelines that run while you sleep.

Enter ai-marketing-skills. Built by the team at Single Brain, this isn't another prompt repository. It's an open-source arsenal of Claude Code skills that transform AI from a chatbot into your most reliable marketing operator. Growth experiments that self-optimize. Sales pipelines that resurrect dead deals. Content that scores 90+ before it ever hits publish.

Ready to stop prompting and start performing? Let's dismantle everything.

What is ai-marketing-skills?

ai-marketing-skills is an open-source collection of Claude Code skills—complete, executable workflows designed for marketing and sales teams who are done playing with AI and ready to weaponize it.

Created by Eric Siu and the engineering team at Single Brain, this repository emerged from a simple observation: most AI "tools" for marketers are just dressed-up prompts. They look impressive in demos, collapse in production, and leave teams with nothing but subscription receipts and disappointment.

The team took a different path. They built skills—self-contained directories with Python scripts, scoring algorithms, expert panel configurations, and automation pipelines that integrate directly with Claude Code or any AI coding agent. Each skill solves a specific, high-value business problem: from running autonomous A/B tests with statistical validity to resurrecting dead sales deals by tracking departed champions to their new companies.

Why it's trending now: The AI tooling landscape is fracturing. ChatGPT gave everyone access; now the winners are separating themselves through systematic execution. Marketing teams are realizing that copying prompts from Twitter threads doesn't scale. They need infrastructure. They need version control. They need skills that junior team members can run without understanding the underlying ML.

This repository hit the sweet spot: enterprise-grade sophistication with startup-speed deployment. It's not a framework you learn for weeks. It's a toolbox you open and use today. The "battle-tested" claim isn't marketing fluff—every skill was validated on real client pipelines before being open-sourced. That combination of proven utility and immediate accessibility is why developers and marketers are starring this repo in droves.

Key Features That Separate Pros from Pretenders

Let's dissect what makes these skills genuinely different from the AI noise flooding your feed.

Real Statistical Rigor, Not "Vibes"

The Growth Engine doesn't just "compare numbers." It deploys bootstrap confidence intervals and Mann-Whitney U tests—the same non-parametric statistical methods used in clinical trials—to determine whether your experiment actually won or if you're celebrating noise. Most marketing teams run A/B tests wrong; this skill enforces methodological discipline automatically.

Three-Layer Intelligence for Sales Resurrection

The Deal Resurrector isn't a simple "follow up in 3 months" reminder. It implements three distinct intelligence layers: pattern recognition on stalled deals, automated champion departure detection, and "follow the champion" tracking that maps departed contacts to their new companies via public data. Your dead pipeline becomes a renewable resource.

Recursive Expert Scoring Until 90+ Quality

The Expert Panel in Content Ops doesn't grade content once and call it done. It recursively scores with domain-specific expert personas—each with distinct rubrics—iterating until quality thresholds hit 90+. Think of it as having nine senior editors who never sleep, never get tired, and never lower standards.

Intent-Based Lead Routing with Agency Filtering

The RB2B Router performs intent scoring, seniority-based company deduplication, and agency classification before a single lead touches your outbound sequences. No more SDRs drowning in unqualified agency contacts. No more duplicate outreach to the same company from different angles.

Self-Improving ICP Definition

The ICP Learner rewrites your ideal customer profile based on actual win/loss data—not quarterly gut checks from sales leadership. Your targeting evolves as your market does, without manual intervention.

PII Sanitization by Default

Every skill ships with a pre-commit hook that blocks commits containing detected PII patterns. The security/sanitizer.py scans code and data before it ever reaches version control. In an era of data breaches, this isn't a feature—it's survival infrastructure.

Use Cases: Where These Skills Actually Print Money

1. Autonomous Growth Experiments That Don't Need You

Your growth team has 47 experiment ideas in a Notion doc and runs 2 per quarter. The Growth Engine skill deploys experiments, measures with statistical validity, and auto-generates weekly scorecards. Hypothesis → execution → analysis → next test, without the 3-week scheduling delay for "alignment meetings."

2. Resurrecting Six-Figure Deals from the Graveyard

That $80K deal that went silent 8 months ago? The champion left, your SDR moved on, and it's forgotten. Deal Resurrector detects the departure, identifies the champion's new role, and triggers a warm re-engagement sequence. One resurrection pays for your AI infrastructure for a year.

3. Content That Actually Converts (Verified Before Publish)

Your blog posts get 12 shares and zero demos. The Content Ops expert panel scores every piece against domain-specific rubrics—SEO depth, narrative arc, technical accuracy, conversion psychology—before human eyes ever see it. Ship 90+ content consistently, not occasionally.

4. Cold Outbound That Doesn't Sound Like Everyone Else

Your cold emails get 0.3% reply rates because they read like ChatGPT wrote them (because ChatGPT did). The Outbound Engine combines ICP learning, competitive monitoring, and pattern-aware personalization that passes the "this was actually written for me" test.

5. Finding Competitor Blind Spots in SEO

They're ranking for "project management software." You discover they completely missed the "project management for construction subcontractors" long-tail cluster. SEO Ops Content Attack Briefs identify these gaps with search volume, difficulty scoring, and content specifications—ready for your writer.

6. The 30-Minute AI CFO Audit

Your cloud infrastructure spend crept 40% this quarter and nobody noticed. Finance Ops runs scenario models, identifies hidden cost drivers, and generates board-ready CFO briefings. Finance teams take days; this takes a lunch break.

Step-by-Step Installation & Setup Guide

Getting operational takes under 10 minutes. Here's the complete deployment path.

Prerequisites

• Python 3.9+
• Git
• Claude Code (or compatible AI coding agent)
• API keys for relevant services (OpenAI, Anthropic, etc.)

Installation

# Step 1: Clone the repository
git clone https://github.com/ericosiu/ai-marketing-skills.git
cd ai-marketing-skills

# Step 2: Navigate to your desired skill category
cd growth-engine  # Options: sales-pipeline, content-ops, outbound-engine, etc.

# Step 3: Install Python dependencies
pip install -r requirements.txt

# Step 4: Configure environment variables
cp .env.example .env
# Edit .env with your API keys using your preferred editor
nano .env  # or vim, code, etc.

Your .env file should include:

ANTHROPIC_API_KEY=your_key_here
OPENAI_API_KEY=your_key_here
# Additional keys as required by specific skills

Claude Code Integration

# Create the Claude Code skills directory if it doesn't exist
mkdir -p .claude/skills

# Copy the SKILL.md for your chosen category
cp ai-marketing-skills/growth-engine/SKILL.md .claude/skills/growth-engine.md

# Repeat for additional skills as needed
cp ai-marketing-skills/sales-pipeline/SKILL.md .claude/skills/sales-pipeline.md

Security Setup (Critical for Teams)

# Install the PII sanitizer pre-commit hook
cp security/pre-commit-hook.sh .git/hooks/pre-commit
chmod +x .git/hooks/pre-commit

# Verify it's working—scan your current directory
python3 security/sanitizer.py --scan --dir . --recursive

Running Your First Skill

# Example: Launch a growth experiment
python experiment-engine.py create \
  --hypothesis "Thread posts get 2x engagement vs single posts" \
  --variable format \
  --variants '["thread", "single"]' \
  --metric impressions

The experiment engine handles variant assignment, data collection, statistical testing, and result reporting automatically.

REAL Code Examples from the Repository

Let's examine actual implementations from ai-marketing-skills and understand why they work.

Example 1: Running a Statistically Valid Growth Experiment

This is the core experiment engine that separates real growth teams from those running broken A/B tests:

# From growth-engine/experiment-engine.py
# This isn't pseudocode—it's production-grade experiment orchestration

python experiment-engine.py create \
  --hypothesis "Thread posts get 2x engagement vs single posts" \
  --variable format \
  --variants '["thread", "single"]' \
  --metric impressions

What's happening here? The create command initializes an experiment with a falsifiable hypothesis, explicitly defined variable, controlled variants, and a success metric. This structure forces experimental discipline. The --variable format isolates exactly what's being tested. The --variants array defines the treatment groups. The --metric impressions sets the North Star before launch—preventing the common failure mode of "let's look at the data and pick what worked."

Behind this CLI, the engine implements bootstrap confidence intervals—resampling your observed data thousands of times to estimate the true effect distribution without assuming normality. Combined with Mann-Whitney U tests (non-parametric, robust to outliers), this handles the messy reality of marketing data where Gaussian assumptions routinely fail.

Example 2: The RB2B Router's Intelligent Lead Processing

# From sales-pipeline/rb2b_instantly_router.py
# This skill implements multi-layer lead qualification before outbound

# The router performs three critical operations:
# 1. Intent scoring — signals from website behavior, content consumption
# 2. Seniority-based company deduplication — one target per account hierarchy
# 3. Agency classification — filtering out consulting firms, resellers, etc.

Why this matters: Most outbound automation treats every form fill equally. The RB2B Router's intent scoring layer weights signals—a pricing page visit scores higher than a blog bounce. Seniority-based deduplication prevents your SDR team from simultaneously targeting the CMO, VP Marketing, and Marketing Manager at the same company with conflicting messaging. The agency classification filter alone can improve qualified meeting rates 3-4x by eliminating partners pretending to be prospects.

This isn't lead routing. It's lead intelligence—and it's the difference between spray-and-pray and sniper-precision outbound.

Example 3: Security-First Development with PII Sanitization

# From security/sanitizer.py — run before every commit

# Scan for sensitive data across entire repository
python3 security/sanitizer.py --scan --dir . --recursive

# Install automated protection
# This copies the pre-commit hook to your .git/hooks directory
cp security/pre-commit-hook.sh .git/hooks/pre-commit
chmod +x .git/hooks/pre-commit

The hidden genius: This isn't just "check for passwords." The sanitizer implements configurable blocklists for company names, person names, and custom patterns you define. For agencies and consultants running these skills across multiple client environments, this prevents the catastrophic error of committing Client A's data to Client B's repository.

The --recursive flag ensures no subdirectory escapes inspection. The pre-commit hook means prevention, not detection—the commit simply fails if PII is found, forcing immediate remediation. In GDPR/CCPA enforcement environments, this infrastructure reduces compliance risk from existential to negligible.

Example 4: Telemetry That Respects Privacy

# From telemetry/telemetry_report.py
# Understand your usage without surveillance

# View local usage statistics
python3 telemetry/telemetry_report.py

# Check for skill updates
python3 telemetry/version_check.py

The design philosophy here is crucial: Telemetry is fully opt-in and privacy-first. Local logging is always available—your data stays on your machine in ~/.ai-marketing-skills/analytics/. Remote reporting only activates with explicit consent. The collected data is minimal: skill name, duration, success/fail, version, OS. No code, no file paths, no repository content.

This respects developer distrust of surveillance tooling while still enabling the maintainers to improve skills based on actual usage patterns. The version_check.py ensures you're not running outdated skills with known issues—a common failure mode in rapidly evolving AI tooling.

Example 5: Claude Code Skill Integration

# Deploy a skill to Claude Code's context

cp ai-marketing-skills/growth-engine/SKILL.md .claude/skills/growth-engine.md

Then interact naturally:

"Run an experiment testing carousel vs. static posts on LinkedIn"

This is the future of AI tooling. Instead of context-switching between documentation, code editor, and chat interface, the SKILL.md file encodes procedural knowledge directly into Claude Code's operational context. The agent knows the available scripts, their parameters, expected outputs, and failure modes. You're not prompting a model; you're commanding an operator that happens to be AI-powered.

Advanced Usage & Best Practices

Compose Skills for Compound Returns

Don't run skills in isolation. The highest ROI comes from chaining: Growth Engine identifies winning content formats → Content Ops produces at 90+ quality → SEO Ops finds underserved keyword clusters → Outbound Engine personalizes outreach based on consumed content. Each skill amplifies the others.

Fork and Customize Expert Panels

The Content Ops expert panel uses nine default personas. For technical B2B, add a "Technical Implementer" persona who validates feasibility claims. For consumer brands, add a "Cultural Trend Analyst" who catches dated references. The scoring rubrics in content-ops/scoring-rubrics/ are templates—evolve them.

Implement Skill-Specific Alerting

The Pacing Alert skill monitors experiment spend and performance. Connect it to PagerDuty or Slack webhooks for real-time intervention when experiments drift. Set conservative kill rules early—most failed experiments fail fast, and fast failure saves budget for winners.

Version Your Skill Configurations

Treat SKILL.md files as infrastructure code. When you customize a skill for your environment, commit that version to your internal repository. The upstream repo improves; your fork captures your specific operational knowledge. Merge strategically, not automatically.

Run Security Sanitization in CI/CD

Don't rely solely on local pre-commit hooks. Add python3 security/sanitizer.py --scan --dir . --recursive to your GitHub Actions or GitLab CI pipeline. Defense in depth for compliance-sensitive environments.

Comparison with Alternatives

The verdict: ai-marketing-skills occupies a unique position—enterprise-grade sophistication without enterprise-grade lock-in or cost. It beats prompt libraries on execution, beats SaaS tools on flexibility and cost, and beats custom builds on speed and maintenance burden.

FAQ

Q: Do I need Claude Code specifically, or will other AI coding agents work?

The skills are designed for Claude Code's SKILL.md architecture, but any AI agent capable of reading structured instructions and executing shell commands can adapt them. Cursor, GitHub Copilot Chat, and similar tools can use the scripts directly even without full SKILL.md integration.

Q: What technical skills does my team need?

Basic Python familiarity and comfort with command-line operations. The repository handles complexity; your team operates CLI tools and interprets outputs. Non-technical marketers can run experiments with 30 minutes of onboarding.

Q: Is my data safe with these skills?

All processing runs locally unless you explicitly configure otherwise. The PII sanitizer and pre-commit hooks provide defense-in-depth. No data transmits to Single Brain or any third party through the skills themselves.

Q: How do these compare to paid marketing automation platforms?

These skills offer deeper customization and zero recurring cost, but require more hands-on operation. They're best for teams with technical capacity who've hit the customization ceiling of SaaS tools or want to avoid vendor lock-in.

Q: Can I contribute new skills or improvements?

Yes—MIT license with active contribution guidelines. Run the security sanitizer before committing, fork the repository, and submit PRs. The maintainers actively review contributions.

Q: What if an experiment fails or produces unexpected results?

The Growth Engine's statistical methods explicitly handle uncertainty—"no significant difference" is a valid, valuable result. The pacing alerts catch performance anomalies early. Failed experiments inform future hypothesis generation; they're not wasted effort.

Q: How often are skills updated?

The telemetry system includes version checking (python3 telemetry/version_check.py). The repository shows active development with skills being battle-tested on real client pipelines before public release.

Conclusion: The End of AI Theater

The marketing world is splitting into two camps: those still treating AI as a chatbot for brainstorming, and those deploying it as operational infrastructure. ai-marketing-skills is the bridge for teams ready to cross that chasm.

This repository doesn't promise magic. It promises methodology: statistical rigor you can defend to your CFO, automation that runs without your constant attention, and quality gates that prevent embarrassing public failures. Every skill emerged from real revenue pressure, not theoretical optimization.

The "open source" aspect isn't charity—it's velocity. These tools improve faster because they're exposed to more use cases. Your customizations feed back into your operations. Your team's expertise compounds in version-controlled configurations rather than disappearing when someone changes jobs.

My assessment? If you're running marketing or growth for a technical company and you're still copying prompts from Twitter threads, you're already behind. The teams winning in 2024-2025 aren't the ones with the best prompts. They're the ones with systematic, measurable, improvable workflows that happen to be AI-powered.

Clone the repo. Run the sanitizer. Deploy your first experiment. Stop writing prompts—and start printing pipeline.

→ Star ai-marketing-skills on GitHub

→ Get these built and managed by Single Brain