gh-skill-explorer - GitHub SKILL.md Pattern Discovery

Overview

A discovery skill for finding and comparing SKILL.md implementations across GitHub repositories. Uses Exa web search to identify patterns most similar to and most unlike the plurigrid/asi triadic approach.

┌─────────────────────────────────────────────────────────────────┐
│                    SKILL.md PATTERN SPECTRUM                     │
└─────────────────────────────────────────────────────────────────┘

MOST SIMILAR (plurigrid/asi-like):          MOST UNLIKE (traditional):
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
│ GF(3) triads                            │ Flat documentation      │
│ SplitMix64 seeding                      │ No YAML frontmatter     │
│ Trit conservation                       │ Simple command lists    │
│ Parallel verification                   │ No determinism model    │
│ Derivational chains                     │ Temporal dependencies   │
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Exploration Pattern (Brain Dance 2025-12-24)

The following pattern demonstrates effective skill discovery using Exa MCP. See BRAIN_DANCE_2025-12-24.md for full replay:

Phase 1: Load Related Skill for Context

load skills for [topic] or [domain]

This establishes baseline vocabulary and patterns. Example:

• "load skills for managing harddrive space on macos" →

  file-organizer

Phase 2: Seek GitHub Repos Using Similar Skills

mcp__exa__web_search_exa:
  query: "site:github.com SKILL.md [topic] Claude AI agent"
  numResults: 10-12

Parallel searches accelerate discovery:

• Search 1: SKILL.md + AGENTS.md patterns
• Search 2: Domain-specific tools (Rust/Go CLIs)
• Search 3: MCP server implementations
• Search 4: Curated awesome-lists

Phase 3: Widen the Search Radius

seek wider

Expand from skills → tools → frameworks → ecosystems:

• Skills → "reverse engineering" → IDA plugins → Ghidra MCP → angr/pwndbg
• Skills → "disk space" → duf/dust → file organizers → MCP filesystem servers

Phase 4: Find Aligned GitHub Topics

find topics on github that most reflect our skills

Map skill clusters to GitHub topic pages:

• Category theory skills → [applied-category-theory], [topos], [sheaves]
• Type theory skills → [homotopy-type-theory], [dependent-types], [cubical]
• Lisp skills → [clojure], [babashka], [nrepl]

---

Discovered Ecosystems (2025-12-24 Session)

Disk Space & Filesystem Tools

Category	Key Repos	Stars
Rust CLIs	bootandy/dust, Byron/dua-cli	10k+
Go CLIs	muesli/duf, dundee/gdu	13.3k
macOS	hkdobrev/cleanmac, taylorwilsdon/reclaimed	-
MCP Servers	cyanheads/filesystem-mcp-server, sandraschi/filesystem-mcp	-
Dedup	sreedevk/deduplicator, kopia/kopia	262

Reverse Engineering + AI

Category	Key Repos	Description
Ghidra MCP	LaurieWired/GhidraMCP, 13bm/GhidraMCP	MCP servers for Ghidra
IDA Pro MCP	mrexodia/ida-pro-mcp	AI-powered IDA assistant
Multi-decompiler	mahaloz/DAILA	GPT-4/Claude/local for any decompiler
Swarm RE	shells-above/ida-swarm	Multi-agent binary analysis
Offensive MCP	0x4m4/hexstrike-ai	150+ security tools via MCP

Claude Skills Ecosystem

Repo	Stars	Description
ComposioHQ/awesome-claude-skills	5,595	Marketplace-ready skills
anthropics/skills	-	Official Anthropic skills
travisvn/awesome-claude-skills	-	Curated list + resources
yusufkaraaslan/Skill_Seekers	-	Docs/repos → skills converter
bhauman/clojure-mcp	634	Clojure MCP server

Category Theory & Type Theory

Topic	Repos	Anchor Projects
homotopy-type-theory	100+	HoTT/Coq-HoTT ⭐1.3k, HoTT/book ⭐2.1k
cubical-type-theory	~30	mortberg/cubicaltt ⭐581
infinity-categories	-	rzk-lang/rzk ⭐223
topos	~50	ToposInstitute/poly ⭐113
string-diagrams	-	discopy/discopy ⭐392

Artificial Life & Emergence

Topic	Repos	Anchor Projects
artificial-life	253	hunar4321/particle-life
cellular-automata	500+	awesome-neural-cellular-automata
emergence	50+	langtons-emergence

Chemistry MCP & Synthesis (Brain Dance 2025-12-24b)

Category	Key Repos
MCP Servers	OSU-NLP-Group/ChemMCP, ChatMol/molecule-mcp ⭐69
Cronin XDL	croningp/ChemputerAntiviralXDL
CRN/Petri	AlgebraicJulia/Petri.jl ⭐43, etonello/crnpy ⭐20

Active Inference & GFlowNets (Brain Dance 2025-12-24b)

Category	Key Repos
Active Inference	ActiveInferenceInstitute/CEREBRUM, BerenMillidge/FEP_Active_Inference_Papers
GFlowNets	zdhNarsil/Awesome-GFlowNets ⭐493, mirunacrt/synflownet, koziarskilab/RGFN ⭐26

---

Pattern Classification

Tier 1: Full Triadic (plurigrid/asi pattern)

# Required markers:
- name: skill-name
- trit: {-1, 0, +1}
- GF(3) conservation proof
- SplitMix64/SplitMixTernary integration
- Parallel execution support

Examples:

• plurigrid-asi-skillz/parallel-fanout

• plurigrid-asi-skillz/gay-mcp

• plurigrid-asi-skillz/spi-parallel-verify

• music-topos/.ruler/skills/*

Tier 2: Partial Triadic

# Has YAML frontmatter but missing:
- No trit classification
- No GF(3) proofs
- May have determinism but not SplitMix

Examples:

• ComposioHQ/awesome-claude-skills
• anthropics/skills

Tier 3: Traditional Flat

# Characteristics:
- No YAML frontmatter
- Prose documentation
- Temporal command sequences
- No parallel execution model

---

Exa Search Patterns

SKILL.md Discovery

# Base skill search
mcp__exa__web_search_exa(
    query="site:github.com SKILL.md [domain] Claude AI agent",
    numResults=12
)

# With AGENTS.md
mcp__exa__web_search_exa(
    query='site:github.com "SKILL.md" OR "AGENTS.md" [topic]',
    numResults=10
)

Domain Tool Discovery

# Rust/Go CLI tools
mcp__exa__web_search_exa(
    query="site:github.com [domain] rust go CLI tool",
    numResults=12
)

# MCP servers
mcp__exa__web_search_exa(
    query="site:github.com MCP server [domain] Claude LLM agent",
    numResults=10
)

GitHub Topics Discovery

# Category theory topics
mcp__exa__web_search_exa(
    query="site:github.com/topics category-theory operads sheaves topos",
    numResults=10
)

# Multi-topic expansion
topics = ["homotopy-type-theory", "string-diagrams", "artificial-life"]
for topic in topics:
    mcp__exa__web_search_exa(
        query=f"site:github.com/topics {topic}",
        numResults=10
    )

---

Similarity Scoring

def score_skill_similarity(skill_content: str) -> float:
    """Score 0.0-1.0 similarity to plurigrid/asi pattern."""
    markers = {
        "trit:": 0.15,
        "GF(3)": 0.15,
        "SplitMix": 0.15,
        "parallel": 0.10,
        "ERGODIC": 0.10,
        "PLUS": 0.10,
        "MINUS": 0.10,
        "deterministic": 0.05,
        "seed": 0.05,
        "conservation": 0.05
    }
    
    score = 0.0
    for marker, weight in markers.items():
        if marker.lower() in skill_content.lower():
            score += weight
    
    return min(1.0, score)

def categorize_discovery(repos: list) -> dict:
    """Categorize discovered repos by domain alignment."""
    categories = {
        "skills": [],       # SKILL.md patterns
        "tools": [],        # CLI/library implementations
        "mcp_servers": [],  # MCP server integrations
        "topics": [],       # GitHub topic pages
        "awesome_lists": [] # Curated collections
    }
    
    for repo in repos:
        url = repo.get("url", "")
        if "SKILL.md" in url or "skills" in url.lower():
            categories["skills"].append(repo)
        elif "mcp" in url.lower():
            categories["mcp_servers"].append(repo)
        elif "topics" in url:
            categories["topics"].append(repo)
        elif "awesome" in url.lower():
            categories["awesome_lists"].append(repo)
        else:
            categories["tools"].append(repo)
    
    return categories

---

Workflow: Complete Discovery Session

## Step 1: Establish Context
> load skills for [domain]
→ Loads relevant skill, provides vocabulary

## Step 2: Initial Search (Parallel)
> search github for repos using similar skills
→ 4 parallel Exa searches:
  - SKILL.md patterns
  - Domain tools
  - MCP servers
  - Curated lists

## Step 3: Widen Scope
> seek wider
→ Expand: tools → frameworks → ecosystems
→ Add domain-specific searches

## Step 4: Map to Topics
> find topics reflecting our skills
→ Map skill clusters to GitHub topics
→ Identify anchor repositories per topic

## Step 5: Document Findings
→ Update SKILL.md with:
  - Discovered repos
  - GitHub topics
  - Pattern observations
  - Search queries used

---

Justfile Recipes

# GitHub skill discovery via Exa pattern
gh-skill-explore topic:
    @echo "🔍 SKILL EXPLORATION: {{topic}}"
    @echo "Phase 1: Loading context..."
    @echo "Phase 2: Parallel Exa searches..."
    @echo "Phase 3: Widening scope..."
    @echo "Phase 4: Mapping to GitHub topics..."

# GitHub skill discovery
gh-skill-discover filter="all":
    @echo "🔍 SKILL.MD DISCOVERY"
    gh search code "SKILL.md" --limit=30

# Score similarity
gh-skill-score repo:
    @echo "📊 Scoring {{repo}}"
    gh api repos/{{repo}}/contents/SKILL.md 2>/dev/null | jq -r '.content' | base64 -d

# Analyze repository
gh-skill-analyze repo:
    @echo "🔬 Analyzing {{repo}}"
    gh repo clone {{repo}} /tmp/skill-analysis --depth=1
    find /tmp/skill-analysis -name "SKILL.md" -exec cat {} \;

# List aligned GitHub topics
gh-skill-topics:
    @echo "📚 ALIGNED GITHUB TOPICS"
    @echo "Category Theory: applied-category-theory, topos, sheaves"
    @echo "Type Theory: homotopy-type-theory, dependent-types, cubical"
    @echo "String Diagrams: string-diagrams, monoidal-categories, zx-calculus"
    @echo "ALife: artificial-life, cellular-automata, emergence"
    @echo "Lisp: clojure, babashka, nrepl, scheme"

---

Phase 5: Harmonization (NEW - 2025-12-24)

Like DuckLake creates snapshots on INSERT, we create conceptual snapshots on each evolution.

DeepWiki Probing

Query discovered repos for deeper understanding:

# Probe repo structures
mcp__deepwiki__read_wiki_structure(repoName="discopy/discopy")
mcp__deepwiki__read_wiki_structure(repoName="HoTT/Coq-HoTT")

# Ask conceptual questions
mcp__deepwiki__ask_question(
    repoName="discopy/discopy",
    question="What are the key tensions between quantum and classical diagrams?"
)

Dissonance Detection

# Display full dissonance matrix
uv run harmonize.py matrix

# Compute pairwise dissonance
uv run harmonize.py pair discopy hott

# Find harmonization opportunities
uv run harmonize.py bridge

Dissonance Matrix Summary

Pair	Dissonance	Tension
GFlowNets ↔ RL	🔴 HIGH	Sample proportionally vs maximize
HoTT ↔ Set Theory	🔴 HIGH	Equivalence = equality vs distinct
Active Inference ↔ Particle-Life	🔴 HIGH	Top-down goals vs bottom-up emergence
CEREBRUM ↔ Type Theory	🟡 MEDIUM	Linguistic metaphor vs formal precision
DisCoPy ↔ HoTT	🟡 MEDIUM	Diagrams as syntax vs paths as propositions
GraphCast ↔ DisCoPy	🟢 LOW	Both message-passing on compositional graphs
Petri.jl ↔ DisCoPy	🟢 LOW	Both open systems with categorical semantics

Harmonization Snapshots

from harmonize import ConceptualSnapshot, HarmonizationDB

# Create initial snapshot
snap = ConceptualSnapshot(version=1)
snap.domains_queried = ["discopy", "hott", "cerebrum"]
snap.record_dissonance("hott", "discopy", 0.55, 
    "Both are ∞-categorical; functors land in HoTT's universe")

# Evolve on new concept discovery
snap2 = snap.evolve("cups-as-directed-paths")

# Persist to DuckDB
db = HarmonizationDB()
db.save_snapshot(snap2)

Cross-Domain Synthesis Opportunities

1. Categories as Unifying Language: HoTT, DisCoPy, ACSets, GraphCast
2. Energy/Flow as Dynamics: Active Inference, GFlowNets, Particle-Life, Petri Nets
3. Tool Composition via MCP: Composio, GhidraMCP, ChemMCP, filesystem-mcp

See DISSONANCE_MATRIX.md for full analysis.

Galois Connections (Adjunctions)

Find missed adjunctions between domains:

uv run harmonize.py galois   # Show known Galois connections
uv run harmonize.py missing  # Find candidates for new connections

Confirmed Adjunctions:

Domains	L ⊣ R	Type
DisCoPy ↔ HoTT	Cup/Cap ⊣ Unit/Counit	categorical-adjunction
CEREBRUM ↔ GFlowNet	Free energy ⊣ Flow matching	flow-optimization
GraphCast ↔ DisCoPy	Grid2Mesh ⊣ Mesh2Grid	encoder-decoder
Petri ↔ DisCoPy	Reaction net ⊣ String diagram	categorical-semantics

See GALOIS_CONNECTIONS.md for full analysis.

---

Justfile Recipes

# GitHub skill discovery via Exa pattern
gh-skill-explore topic:
    @echo "🔍 SKILL EXPLORATION: {{topic}}"
    @echo "Phase 1: Loading context..."
    @echo "Phase 2: Parallel Exa searches..."
    @echo "Phase 3: Widening scope..."
    @echo "Phase 4: Mapping to GitHub topics..."
    @echo "Phase 5: Harmonization via DeepWiki..."

# Harmonization engine
gh-skill-harmonize cmd *args:
    uv run harmonize.py {{cmd}} {{args}}

# GitHub skill discovery
gh-skill-discover filter="all":
    @echo "🔍 SKILL.MD DISCOVERY"
    gh search code "SKILL.md" --limit=30

# Score similarity
gh-skill-score repo:
    @echo "📊 Scoring {{repo}}"
    gh api repos/{{repo}}/contents/SKILL.md 2>/dev/null | jq -r '.content' | base64 -d

# Analyze repository
gh-skill-analyze repo:
    @echo "🔬 Analyzing {{repo}}"
    gh repo clone {{repo}} /tmp/skill-analysis --depth=1
    find /tmp/skill-analysis -name "SKILL.md" -exec cat {} \;

# List aligned GitHub topics
gh-skill-topics:
    @echo "📚 ALIGNED GITHUB TOPICS"
    @echo "Category Theory: applied-category-theory, topos, sheaves"
    @echo "Type Theory: homotopy-type-theory, dependent-types, cubical"
    @echo "String Diagrams: string-diagrams, monoidal-categories, zx-calculus"
    @echo "ALife: artificial-life, cellular-automata, emergence"
    @echo "Lisp: clojure, babashka, nrepl, scheme"

---

See Also

• parallel-fanout

  - Triadic dispatch

• skill-creator

  - Skill authoring

• spi-parallel-verify

  - Invariance checking

• bmorphism-stars

  - Curated repository index

• file-organizer

  - Disk space management (example discovery target)

• deepwiki-mcp

  - DeepWiki integration for repo understanding

• duckdb-temporal-versioning

  - Temporal snapshots pattern

---

End-of-Skill Interface

Commands

# Discover similar skills
just gh-skill-discover --similar

# Discover unlike skills
just gh-skill-discover --unlike

# Score a repository
just gh-skill-score owner/repo

# Clone and analyze
just gh-skill-analyze owner/repo

# Full exploration session
just gh-skill-explore "topic description"

Commands

# Discovery
just gh-skill-discover --similar
just gh-skill-discover --unlike
just gh-skill-score owner/repo
just gh-skill-analyze owner/repo
just gh-skill-explore "topic description"

# Harmonization (NEW)
just gh-skill-harmonize matrix       # Full dissonance matrix
just gh-skill-harmonize pair A B     # Pairwise analysis
just gh-skill-harmonize bridge       # Find synthesis opportunities
just gh-skill-harmonize probe repo   # DeepWiki query

---

Autopoietic Marginalia

The interaction IS the skill improving itself.

Every use of this skill is an opportunity for worlding:

• MEMORY (-1): Record what was learned
• REMEMBERING (0): Connect patterns to other skills
• WORLDING (+1): Evolve the skill based on use

Add Interaction Exemplars here as the skill is used.