Intelligence Adjacent: Democratizing Expertise Through AI

What if decades of expertise could be accessible to anyone, guided by AI, through orchestration and scaffolding?

That's the promise of Intelligence Adjacent (IA) - an AI infrastructure framework that augments human intelligence rather than attempting to replace it.

The Problem: Expertise Inaccessibility

Expertise is locked behind years of experience. Want to conduct a professional penetration test? You need years of security training. Need to navigate complex legal compliance? You'll need a law degree and bar admission. Career advancement? You're competing against decades of industry knowledge.

The traditional solution: Hire expensive experts or spend years building expertise yourself.

The AI-powered alternative: Use LLMs to automate everything and hope for the best.

The Intelligence Adjacent approach: Build systems that combine human decision-making with AI-assisted workflows, making expertise accessible while maintaining professional standards.

Why This Matters: Democratizing Expertise

I'm not building this so people have to pay ME for expertise.

I'm building it so they can develop expertise THEMSELVES—guided by AI, through proven workflows.

For too long, expertise has been gatekept behind expensive consultants, years of training, and corporate budgets. Intelligence Adjacent challenges that.

Why should small nonprofits go without security because they can't afford $200/hour consultants?

Why should individuals wait for corporate permission to learn advanced skills?

This framework makes decades of methodology accessible to anyone willing to learn. The goal isn't replacing experts—it's empowering more people to become experts through AI-assisted scaffolding.

Think of it like this: A library doesn't replace readers. It organizes knowledge so more people can access it. That's what Intelligence Adjacent does for professional expertise.

What Intelligence Adjacent Is

Intelligence Adjacent is a production-ready AI infrastructure framework that provides:

1. Specialized Agents - Domain experts that load complete workflows on demand (director, security, writer, advisor, legal)
2. Reusable Skills - Complete methodologies, templates, and processes across security, content, advisory, legal, and infrastructure domains
3. Multiple Engagement Modes - Choose your approach: Director (production), Mentor (learning), or Demo (testing)
4. Multi-Model Architecture - Right model for the right task (Claude Sonnet/Haiku, Grok, Perplexity)
5. VPS Security Tools - Professional-grade tools accessible via AI (Kali pentest, Web3 security, mobile security, REAPER)

Think of skills-based context management like a professional library: Instead of carrying every book you might need, you have a card catalog that loads the exact reference material when you need it. Skills are the books, agents are the librarians who know which books to pull.

The framework runs on Claude Code with this skills-based approach, treating text as a thought primitive - a fundamental building block of thinking.

High-Level Architecture: Why /skills, /agents, /servers?

The Three-Layer Design

Layer 1: /skills/ - The Intelligence

Skills contain complete workflows, methodologies, templates, and tested processes. They're the actual expertise.

Why separate from agents? One skill can serve multiple agents and modes. Update the skill once, and every agent using it improves.

Example: skills/security-testing/SKILL.md contains 3 complete testing methodologies (penetration testing, vulnerability scanning, network segmentation). The security agent loads this skill and auto-detects which mode to use based on your request.

Layer 2: /agents/ - The Thin Wrappers

Agents are lightweight identity layers that load skills and execute workflows. They don't contain intelligence - they orchestrate it.

Why this works: Instead of fragmented specialized agents with duplicated workflows, we have unified multi-skill agents that dynamically load expertise.

Current agents:

• security - Testing + Advisory (Security skills across pentesting, code review, architecture, and compliance)
• writer - Blog + Technical + Reports (Content skills across blog posts, technical documentation, and security reports)
• advisor - Career + Research + QA (Advisory skills across career development, research, and quality assurance)
• legal - Compliance + Risk + Jurisdictional (Legal compliance analysis and risk assessment)
• director - Orchestration (Pure routing layer, no skills)

Layer 3: /servers/ - The Force Multipliers

VPS Code API wrappers that provide direct access to professional security tools via SSH + docker exec pattern.

Why VPS deployment? 70-95% token reduction (meaning 70-95% lower API costs and faster response times) (measured average: 82%) token reduction compared to traditional MCP protocol layers. Tools execute on remote VPS, results stored locally, agents parse efficiently.

Example: Instead of sending 50KB of nmap output through an MCP protocol layer, the wrapper executes ssh vps 'docker exec kali nmap [target]', saves output to local file, and returns a 500-byte summary. When full results are needed, agents read the local file.

Tool categories:

• Kali Pentest - nmap, nuclei, sqlmap, metasploit, etc.
• Web3 Security - slither, mythril, semgrep, etc.
• Mobile Security - apktool, jadx, frida, etc.
• REAPER - Database-driven traffic analysis

Three Engagement Modes: Choose Your Approach

Director Mode (Production) - Fast execution, professional deliverables

• Full workflow with OSINT, threat intel, comprehensive testing
• Duration: 4-6 hours for penetration test
• Best for: Client work, actual engagements, production deliverables

Mentor Mode (Learning) - Same results + skill building

• After each phase: "Want to try this yourself, watch me, or skip teaching?"
• Duration: ~20-30% longer than Director mode
• Best for: Learning, building expertise, progressive skill development

Demo Mode (Testing) - Rapid validation in 5-30 minutes

• Skips OSINT (30-60 min), threat intel (2-3 hours), test planning (15-30 min)
• Safe test targets only (scanme.nmap.org, httpbin.org, localhost)
• Duration: 5-30 minutes for targeted testing
• Time savings: 90-95% reduction vs. full workflow
• Best for: Tool connectivity testing, infrastructure validation, troubleshooting

The value: Same underlying skill, different execution modes. Choose based on your goal.

Multi-Model Architecture: Right Model, Right Task

Claude Sonnet 4.5 (**claude-sonnet-4-5-20250929**) - Complex Analysis

• Latest frontier model from Anthropic (as of January 2025)
• Penetration testing with exploitation
• Legal analysis and compliance review
• Blog writing and content creation
• Personal coaching and career development
• Technical documentation

Claude Haiku 4.5 (**claude-haiku-4-5-20250929**) - Fast Routing

• Fast, efficient model for structured tasks
• Director agent orchestration
• Template-based workflows
• Routine vulnerability scans
• Simple file operations

Grok AI (via OpenRouter) - QA Verification

• x-ai/grok-code-fast-1 - Technical/code content QA review
• x-ai/grok-4-fast - General content QA verification
• Critical use case: Automated blog post QA, legal citation verification
• Note: Hero images via manual Grok UI workflow (no automated generation)

Perplexity Pro (via API) - Research with Citations

• Deep research with automatic citations
• Systematic web research
• Use cases: Career intel, threat intelligence, industry analysis

Why multiple models? Cost optimization + specialized capabilities. Use Haiku for routing ($0.25/MTok input), Sonnet for complex analysis ($3/MTok input), Grok for verification, Perplexity for research.

Complete Skills Inventory

Security Skills:

1. security-testing - Pentest, vuln scan, segmentation testing
2. cybersecurity-advisory - Risk assessments, security guidance
3. code-review - Security-focused code analysis
4. secure-config - Infrastructure hardening
5. benchmark-generation - CIS compliance automation
6. architecture-review - Security architecture assessment
7. dependency-audit - Supply chain security
8. threat-intel - Threat intelligence research

Content Creation Skills:

9. blog-writer - Blog content and news aggregation
10. technical-writing - System documentation and API docs
11. report-generation - Business deliverables and executive reports

Advisory & Research Skills:

12. personal-development - Career advancement, CliftonStrengths coaching, mentorship
13. osint-research - Multi-source research with citations
14. qa-review - Quality assurance and validation

Legal Compliance Skills:

15. legal-compliance - Contract review, CFAA compliance, jurisdictional research

• Critical: NOT legal advice, provides legal information only
• Grok verification: ALL case citations verified before presentation

Infrastructure/System Skills:

16. infrastructure-ops - VPS management, deployments, monitoring
17. create-skill - Skill/agent creation templates
18. blog-workflow - Blog content automation and publishing workflows

Framework Discovery: Zero-Maintenance Scaling

All components use YAML manifests for automatic discovery. No hardcoded registries, no maintenance overhead.

Pattern: Follow template → Drop in location → System finds automatically

7 discoverable types: Skills, Agents, Server Tools, Commands, Hooks, Tools, Workflows

Result: Add a new skill by creating skills/new-skill/MANIFEST.yaml and SKILL.md. The framework automatically discovers it. No updates to CLAUDE.md, no agent modifications, no registry management.

Performance Metrics

Note: All metrics below are based on internal testing and production use of this framework. Your results may vary based on task complexity and infrastructure.

70-95% token reduction (measured average: 82%) - VPS Code API pattern vs. MCP protocol layer

• Traditional: 50KB nmap output through protocol → 50,000 tokens
• VPS pattern: SSH → docker exec → local file → 500 token summary
• Full results available by reading local file when needed
• Impact: Lower API costs, faster response times, same quality results

90-95% Time Savings - Demo mode vs. full workflow

• Full pentest workflow: 4-6 hours (OSINT, threat intel, planning, testing)
• Demo mode: 5-30 minutes (skip OSINT/planning, safe targets only)
• Use case: "Is nuclei working on my VPS?" vs. "Complete vulnerability scan"
• Caveat: Demo mode is for validation only, not production testing

20-30% Learning Overhead - Mentor mode vs. Director mode

• Same deliverables, 20-30% more time
• Progressive skill building through scaffolded learning
• Value: Learn while producing professional-grade work
• ROI: Long-term skill development vs. short-term speed

55% Agent Reduction - From 11 original agents to 5

• Eliminated duplication through skills-based architecture
• One skill serves multiple agents/modes
• Simplified routing via auto-detection
• Maintenance benefit: Update once in /skills, improve everywhere

The Philosophy: Intelligence Adjacent

Text is a thought primitive - a fundamental building block of thinking.

By organizing knowledge into accessible, versionable files (skills) and providing intelligent agents to navigate that knowledge, we create systems that enhance human decision-making rather than attempting to replicate it.

The goal: Upgrade human capabilities, not replace them.

The method: Orchestration and scaffolding over raw intelligence.

The result: Decades of expertise, accessible through AI-assisted workflows, maintaining professional standards.

How You Can Use This

This isn't a commercial product you pay for—it's an approach you can build yourself.

If you want to build something similar:

1. Start small - Pick one domain you know well (security testing, content writing, legal research)
2. Document your workflow - Turn your expertise into step-by-step markdown files
3. Create agent prompts - Wrap your workflows in AI agent prompts that load them on demand
4. Test with real tasks - Use it for actual work, not just demos
5. Iterate based on results - Refine workflows based on what works

The goal isn't copying this exact architecture—it's understanding the pattern: Organize expertise → Make it AI-accessible → Empower others to use it.

Follow the blog at notchrisgroves.com for detailed build guides and deployment tutorials.

What's Next

This framework is in active development with new skills, agents, and capabilities added regularly.

Current focus:

• Legal advisor agent (compliance review, CFAA analysis, jurisdictional research)
• Universal mentor mode expansion (blog writer, advisor skills)
• Additional VPS security tools deployment
• Framework documentation and knowledge base

Future directions:

• Additional specialized agents as clear use cases emerge
• More skills-based workflows across different domains
• Enhanced multi-model orchestration
• Community contributions and skill templates

Built With

• Claude Code - Primary development environment
• Claude 4.5 Sonnet/Haiku - Core AI models
• Grok AI - Verification and specialization
• Perplexity Pro - Research and automation
• OVHcloud VPS - Security tools deployment (Comprehensive tool suite across Kali, Web3, mobile, and exploitation frameworks)
• Hostinger VPS - Production automation (n8n, Vaultwarden, Ollama)
• Twingate - Zero-trust network access

Learn More

• Blog: notchrisgroves.com
• Repository: Intelligence Adjacent Framework
• Documentation: Framework documentation included in repository + blog deployment guides

Intelligence Adjacent: Building systems that augment human intelligence, not replace it.