Skill Maker: AI Skill Factory for Tools

Skill Maker: AI Skill Factory

Version: 1.0.0 Status: Production Ready Trit Assignment: 0 (neutral, structural - scaffolds other skills) Principle: Template-driven skill generation with SplitMix seeding

Purpose

The Skill Maker is a meta-skill that automatically generates domain-specific AI skills by:

1. Analyzing Tool Documentation - Firecrawl documentation, README, API specs
2. Applying SKILL.md Pattern - Deterministic output, ternary polarity, SPI guarantee, parallelism
3. Generating Custom SKILL.md - Fully functional, production-ready skill
4. Registering with Claude Code - Immediate availability as new skill

Architecture

Skill Maker Pipeline
├─ Phase 1: Tool Discovery (Firecrawl + documentation analysis)
├─ Phase 2: Pattern Recognition (extract key operations, outputs, semantics)
├─ Phase 3: SplitMix Adaptation (add deterministic seeding to tool)
├─ Phase 4: Ternary Mapping (classify outputs to GF(3) = {-1, 0, +1})
├─ Phase 5: Parallelism Design (split-stream, work-stealing architecture)
├─ Phase 6: SKILL.md Generation (template expansion with tool specifics)
└─ Phase 7: MCP Integration (register and deploy skill)

Phase 1: Tool Discovery

Firecrawl-Based Documentation Analysis

from firecrawl import FirecrawlApp
import re

class ToolDiscovery:
    def __init__(self, tool_name: str):
        self.tool_name = tool_name
        self.app = FirecrawlApp(api_key=os.getenv("FIRECRAWL_API_KEY"))

    def discover_tool(self) -> dict:
        """Automatically discover tool via web search and documentation."""
        # Search for tool
        search_results = self.app.search(
            f"{self.tool_name} documentation API reference"
        )

        # Extract key URLs
        urls = [result['url'] for result in search_results[:5]]

        # Scrape and analyze
        tool_spec = {
            "name": self.tool_name,
            "description": "",
            "operations": [],
            "inputs": [],
            "outputs": [],
            "examples": [],
            "urls": urls
        }

        for url in urls:
            doc = self.app.scrape_with_markdown(url)
            content = doc['markdown']

            # Extract operations
            operations = self._extract_operations(content)
            tool_spec["operations"].extend(operations)

            # Extract examples
            examples = self._extract_examples(content)
            tool_spec["examples"].extend(examples)

            # Extract description
            if not tool_spec["description"]:
                tool_spec["description"] = content.split('\n')[0:5]

        return tool_spec

    def _extract_operations(self, content: str) -> list:
        """Extract operation names from documentation."""
        # Look for function/method signatures
        patterns = [
            r'def\s+(\w+)\(',           # Python
            r'fn\s+(\w+)\(',            # Rust
            r'function\s+(\w+)\(',      # JavaScript
            r'public\s+\w+\s+(\w+)\(',  # Java
        ]

        operations = []
        for pattern in patterns:
            matches = re.findall(pattern, content)
            operations.extend(matches)

        return list(set(operations))

    def _extract_examples(self, content: str) -> list:
        """Extract code examples from documentation."""
        # Look for code blocks
        examples = re.findall(r'```[\w]*\n(.*?)```', content, re.DOTALL)
        return examples[:5]  # Top 5 examples