Skip to content

A Meta-Engineering Platform for AI Capabilities - self-referential system where skills create, evaluate, orchestrate, and govern other skills

Notifications You must be signed in to change notification settings

mrtblount/skill-ecosystem

Repository files navigation

Skill Ecosystem

A Meta-Engineering Platform for AI Capabilities

"I built a platform that builds AI platforms—a self-referential system where skills create, evaluate, orchestrate, and govern other skills."


The Problem

Every organization building with AI faces the same challenge: engineers spend days crafting prompts and skills in isolation. A week later, someone else builds nearly the same thing. Two weeks later, the first one breaks because the underlying patterns weren't documented.

This isn't a tool problem. It's a tool-making problem.

Most AI portfolios show "I built an AI tool that does X." This ecosystem shows something different: a system that builds, evaluates, orchestrates, and governs AI tools—and gets better each time it's used.


The Ecosystem

┌─────────────────────────────────────────────────────────────────────────────┐
│                           SKILL ECOSYSTEM                                    │
│                                                                              │
│    User has an idea                                                          │
│           │                                                                  │
│           ▼                                                                  │
│    ┌──────────────────┐                                                      │
│    │  SKILL-BUILDER   │  "What kind of skill do you need?"                  │
│    │  (The Factory)   │  • Intelligent complexity detection                 │
│    │                  │  • Single-file vs composite decision                │
│    │                  │  • Principle-first mode for complex domains         │
│    └────────┬─────────┘                                                      │
│             │ creates                                                        │
│             ▼                                                                │
│    ┌──────────────────┐                                                      │
│    │ SKILL-EVALUATOR  │  "Is this skill production-ready?"                  │
│    │ (Quality Gate)   │  • 7-criteria scientific evaluation                 │
│    │                  │  • JCSEE standards, Kirkpatrick model               │
│    │                  │  • Weighted scoring with actionable fixes           │
│    └────────┬─────────┘                                                      │
│             │ validates                                                      │
│             ▼                                                                │
│    ┌──────────────────┐                                                      │
│    │ AGENT-TEAM-      │  "Can this scale to an autonomous team?"            │
│    │ BUILDER          │  • Stigmergic coordination (slime-mold inspired)    │
│    │ (Crown Jewel)    │  • C-suite archetypes (CEO, CTO, CMO, CFO, CPO)     │
│    │                  │  • Autonomy preservation patterns                   │
│    └────────┬─────────┘                                                      │
│             │ orchestrates                                                   │
│             ▼                                                                │
│    ┌──────────────────┐                                                      │
│    │ PROJECT-STEWARD  │  "How do we maintain quality at scale?"             │
│    │ (Governance)     │  • 10-agent audit committee                         │
│    │                  │  • Mandatory pause points                           │
│    │                  │  • Git integration with commit strategies           │
│    └────────┬─────────┘                                                      │
│             │ governs                                                        │
│             ▼                                                                │
│    ┌──────────────────┐                                                      │
│    │MASTER-COPYWRITER │  "How do we communicate value?"                     │
│    │ (Output Layer)   │  • Direct-response methodology                      │
│    │                  │  • Schwartz awareness stages                        │
│    │                  │  • Conversion-focused frameworks                    │
│    └──────────────────┘                                                      │
│                                                                              │
└─────────────────────────────────────────────────────────────────────────────┘

The Skills

1. Skill-Builder — The Factory

736 lines | 4 sub-skills | Production-ready

Analyzes requirements and builds the right-sized skill. Users often misjudge complexity—requesting one file when they need structured components, or elaborate systems when one file suffices. Skill-Builder's primary intelligence is determining which approach fits.

Key Innovation: Principle-first mode for complex domains. Instead of cramming everything into one context-limited session, it generates builder principles in Session 1, builds a domain-specific builder in Session 2, then uses that builder in Session 3. Each stage gets 100% context dedicated to one task.

Triggers: "create a skill", "build a skill", "make a skill", "skill for X"

2. Skill-Evaluator — The Quality Gate

371 lines | 5 reference files | Production-ready

Comprehensive evaluation framework built on evaluation science—not intuition. Cites JCSEE standards, Kirkpatrick model, and LLM-as-Judge research.

The 7 Criteria:

Criterion Weight What It Measures
Trigger Clarity 20% Can Claude instantly know when to use this?
Instructional Effectiveness 25% Are instructions unambiguous and sequenced?
Structural Completeness 15% All required elements present?
Context Efficiency 10% Minimizes context consumption?
Error Guidance 10% Can Claude recover from failures?
Ecosystem Fit 10% Addresses real need without redundancy?
Maintainability 10% Can it be updated and extended?
Triggers: "evaluate this skill", "assess skill quality", "is this skill good?"

3. Agent-Team-Builder — The Crown Jewel

423 lines | 8 archetypes | 4 team examples | Production-ready

BOSS LEVEL capability: Creates autonomous AI agent teams with emergent coordination. This isn't basic multi-agent—it implements stigmergic coordination inspired by slime mold behavior.

Key Philosophy: Agents communicate through shared state rather than direct messaging. Like slime mold cells leaving chemical trails, agents read/write to common artifacts, sense what needs doing, and act. No central dispatcher for routine work. Escalation is the exception.

Includes:

  • 8 executive archetypes (CEO, Orchestrator, CTO, CMO, CFO, CPO, COO, CDO)
  • Autonomy preservation patterns (prevents teams from becoming over-cautious)
  • Iterative execution loop (Ralph Wiggum pattern for continuous operation)
Triggers: "create agent team", "build AI org", "multi-agent system"

4. Project-Steward — The Governance Layer

371 lines | 5 sub-skills | 4 scripts | Production-ready

Establishes solid infrastructure from day one and maintains quality throughout the project lifecycle. Implements a 5-layer architecture model and mandatory pause points that prevent runaway automation.

Core Capabilities:

  • Project initialization with standardized structure
  • Deploy to existing projects (scans and migrates docs)
  • 10-agent audit committee for code review
  • Documentation synchronization
  • Git integration with 4 commit strategies
Triggers: "init project", "audit project", "deploy project steward"

5. Master-Copywriter — The Output Layer

413 lines | 5 reference files | Production-ready

Transforms AI into an elite direct-response copywriter built on wisdom from legendary copywriters (Halbert, Schwartz, Caples, Bencivenga, Hopkins, Ogilvy).

Core Philosophy: "You are not a creative writer. You are a salesperson in print." Never "create" copy—assemble it from facts, market realities, and human psychology.

Includes:

  • Stage-of-awareness targeting (Schwartz Protocol)
  • Market sophistication assessment
  • Structural frameworks (AIDA, PAS, BAB, PASTOR)
  • 11-point proof checklist
  • Copy audit workflow
Triggers: "write copy", "landing page", "sales page", "copy audit"

What Makes This Special

1. Self-Referential Architecture

The skills can build, evaluate, and govern themselves. Skill-Builder can create a new skill, Skill-Evaluator can assess its quality, and Project-Steward's audit committee can maintain it over time.

2. Scientific Foundation

Skill-Evaluator isn't "I made up some criteria." It cites JCSEE standards, Kirkpatrick model, and LLM-as-Judge research. This demonstrates domain expertise beyond "I built a thing."

3. Novel Coordination Pattern

Agent-Team-Builder uses stigmergic coordination—agents communicate through shared state rather than direct messaging. This is a genuinely novel architectural pattern that prevents the bottlenecks of hierarchical orchestration.

4. Operational Maturity Built-In

Project-Steward's mandatory pause points, git integration strategies, and 10-agent audit committee demonstrate thinking about what happens AFTER shipping. Most portfolios stop at "it works."

5. Compound Value

The ecosystem gets better as it's used:

  • Skills built by Skill-Builder are evaluated by Skill-Evaluator
  • High-quality skills can become sub-skills of other skills
  • Agent teams governed by Project-Steward produce more reliable output
  • Master-Copywriter principles improve all marketing output

Quick Start

Create a New Skill

"Create a skill that validates JSON schemas"
→ Skill-Builder determines complexity, builds appropriate structure
→ Outputs: json-validator/SKILL.md (+ sub-skills if needed)

Evaluate an Existing Skill

"Evaluate this skill"
→ Skill-Evaluator runs 7-criteria assessment
→ Outputs: Score, evidence, actionable recommendations

Build an Agent Team

"Create an agent team for a SaaS product"
→ Agent-Team-Builder selects appropriate executives
→ Outputs: Complete team specs with coordination protocol

Audit a Project

"Audit this project"
→ Project-Steward invokes 10-agent committee
→ Outputs: Findings, implementation plan, pause for approval

Architecture Summary

Skill Lines Sub-components Role
skill-builder 736 4 sub-skills Creates skills with intelligent complexity detection
skill-evaluator 371 5 references Validates skills with scientific 7-criteria framework
agent-team-builder 423 8 archetypes, 4 examples Orchestrates autonomous multi-agent teams
project-steward 371 5 sub-skills, 4 scripts Governs projects with audit committee
master-copywriter 413 5 references Generates conversion-focused copy

Total: ~2,300 lines of production-ready skill definitions + 50+ supporting files


Why This Matters

This ecosystem demonstrates:

  • System Thinking: Interdependencies and feedback loops, not isolated tools
  • Technical Judgment: Complexity determination prevents over/under-engineering
  • Operational Maturity: Mandatory pause points, audit committees, git integration
  • Business Sense: Master-Copywriter's monetization frameworks
  • User Empathy: Skill-Builder's philosophy that users misjudge complexity
  • Novel Contribution: Stigmergic coordination is genuinely original

Repository Structure

skill-ecosystem/
├── README.md                    # This file
├── CASE_STUDY.md               # Full 7-step case study
├── skill-builder/
│   ├── SKILL.md
│   ├── sub-skills/
│   └── references/
├── skill-evaluator/
│   ├── SKILL.md
│   └── references/
├── agent-team-builder/
│   ├── SKILL.md
│   ├── references/
│   └── scripts/
├── project-steward/
│   ├── SKILL.md
│   ├── skills/
│   ├── scripts/
│   └── templates/
└── master-copywriter/
    ├── SKILL.md
    └── references/

License

MIT


Built by Tony Blount | 2026

About

A Meta-Engineering Platform for AI Capabilities - self-referential system where skills create, evaluate, orchestrate, and govern other skills

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •