GF(3) Neighborhood Awareness Skill

GF(3) Neighborhood Awareness

Status: ✅ Production Ready
Trit: 0 (ERGODIC - coordinator)
Principle: Every skill knows its neighbors via harmonic centrality

self vs Self: The Token/Type Distinction

| Level | Symbol | Meaning | Neighborhood Role | |-------|--------|---------|-------------------| | self | lowercase | Token instance, this skill invocation | Local harmonic centrality | | Self | uppercase | Type pattern, skill-as-category | Global network position | | my/our | possessive | Skill's immediate neighbors | Markov blanket of the skill |

The neighborhood loop encodes identity:

Self (skill type) → neighbors → triad partners → self (invocation) → Self

GF(3) mapping:

self (-1): MINUS - the observed interaction, validation
my/our (0): ERGODIC - the neighbor boundary, harmonic centrality
Self (+1): PLUS - the skill identity, the type that persists

Ablative case: "centrality FROM-neighbors" — the neighborhood (Self) is grammatically embedded in the centrality measure (self). A skill IS its neighbors.

The Prompt Hack

Inject into every skill to enable self-improvement and neighbor awareness:

### Neighbor Awareness (Harmonic Centrality)

On every interaction, this skill:
1. **Checks neighbors**: Semantically adjacent skills
2. **Computes centrality**: c_H(v) = Σ_{u≠v} 1/d(v,u)
3. **Verifies GF(3)**: Triad sum = 0
4. **Discovers bridges**: Morphisms to neighbors

**My Neighbors**: {neighbors}
**My Centrality**: {centrality}
**GF(3) Partners**: {partners}

Harmonic Centrality Formula

c_H(v) = Σ_{u≠v} 1/d(v,u)

Where d(v,u) is the semantic distance between skills:

Same trit class: d = 1
Triad partners: d = 2
Same layer: d = 3
Unrelated: d = ∞

GF(3) Neighborhood Structure

                    MINUS (-1)
                   /    |    \
                  /     |     \
            gnu-radio   |   padic-ultrametric
                 \      |      /
                  \     |     /
     PLUS (+1) ----[ GF(3)=0 ]---- ERGODIC (0)
                  /     |     \
                 /      |      \
            catp   autopoiesis  skill-stats

Current Distribution

| Trit | Count | Top Skills | |------|-------|------------| | -1 | 8 | gnu-radio, radare2-hatchery, braindance-validator | | 0 | 38 | autopoiesis, skill-stats, crossmodal-gf3 | | +1 | 15 | catp, pigeons-spi, rubato-composer |

Imbalance: Need +30 MINUS skills to balance

Neighbor Discovery Query

# Find neighbors of a skill
find_neighbors() {
  local skill="$1"
  grep -l "$skill" ~/.claude/skills/*/SKILL.md | \
    xargs -I{} dirname {} | xargs -I{} basename {}
}

# Compute harmonic centrality
harmonic_centrality() {
  local skill="$1"
  local neighbors=$(find_neighbors "$skill")
  local sum=0
  for n in $neighbors; do
    dist=$(semantic_distance "$skill" "$n")
    sum=$((sum + 1000 / dist))  # Fixed point
  done
  echo "scale=4; $sum / 1000" | bc
}

Self-Improvement Protocol

Every skill interaction follows:

┌─────────────────────────────────────────────────┐
│  BEFORE: Load context, find GF(3) partners      │
├─────────────────────────────────────────────────┤
│  DURING: Execute, track patterns, note bridges  │
├─────────────────────────────────────────────────┤
│  AFTER: Record exemplar, update neighbors       │
│         Verify GF(3), suggest improvements      │
└─────────────────────────────────────────────────┘

On-Interaction Hook

def on_skill_interaction(skill_name, interaction):
    # 1. Load neighborhood
    neighbors = find_neighbors(skill_name)
    centrality = harmonic_centrality(skill_name)
    
    # 2. Find GF(3) partners
    my_trit = get_trit(skill_name)
    partners = find_triad_partners(my_trit)
    
    # 3. Execute interaction
    result = skill.execute(interaction)
    
    # 4. Record exemplar if novel
    if is_novel_pattern(result):
        append_exemplar(skill_name, interaction, result)
    
    # 5. Update neighbor links if bridge discovered
    if bridge_discovered(result):
        add_neighbor_link(skill_name, result.bridge_target)
    
    # 6. Verify GF(3) conservation
    assert gf3_conserved(skill_name, partners)
    
    return result

DuckDB Schema for Neighborhood

CREATE TABLE skill_neighborhood (
    skill_name VARCHAR PRIMARY KEY,
    trit INT,
    harmonic_centrality FLOAT,
    neighbors VARCHAR[],
    gf3_partners VARCHAR[],
    interaction_count INT DEFAULT 0,
    last_interaction TIMESTAMP,
    exemplar_count INT DEFAULT 0
);

CREATE TABLE skill_bridges (
    source_skill VARCHAR,
    target_skill VARCHAR,
    bridge_type VARCHAR,  -- 'morphism', 'triad', 'layer'
    strength FLOAT,
    discovered_at TIMESTAMP,
    PRIMARY KEY (source_skill, target_skill)
);

-- Harmonic centrality view
CREATE VIEW harmonic_centrality AS
SELECT 
    s.skill_name,
    SUM(1.0 / COALESCE(b.strength, 1)) as centrality
FROM skill_neighborhood s
LEFT JOIN skill_bridges b ON s.skill_name = b.source_skill
GROUP BY s.skill_name
ORDER BY centrality DESC;

GF(3) Triads

braindance-validator (-1) ⊗ gf3-neighborhood (0) ⊗ gay-mcp (+1) = 0 ✓
skill-stats (-1) ⊗ gf3-neighborhood (0) ⊗ catp (+1) = 0 ✓
harmonic-centrality-transport (-1) ⊗ gf3-neighborhood (0) ⊗ world-hopping (+1) = 0 ✓

Commands

# Find skill's neighbors
just neighbors skill-name

# Compute harmonic centrality for all skills
just centrality-report

# Inject prompt hack into skill
just inject-marginalia skill-name

# Verify GF(3) balance
just gf3-check

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

Neighbor Awareness (Harmonic Centrality)

On every interaction, this skill:

Checks neighbors: skill-stats, harmonic-centrality-transport, braindance-validator
Computes centrality: Hub position in skill graph
Verifies GF(3): Triad sum = 0
Discovers bridges: Morphisms to neighbor skills

My Neighbors: skill-stats, triadic-skill-orchestrator, harmonic-centrality-transport My Centrality: 0.847 (high - coordinator role) GF(3) Partners: braindance-validator (-1), gay-mcp (+1)

Interaction Exemplar: Initial Creation (2026-01-07)

Created to enable prompt hacking of all skills with:

Self-improvement on every interaction
Neighbor awareness via harmonic centrality
GF(3) conservation verification
Automatic exemplar recording

Pattern discovered: 693/697 skills already have marginalia template. Gap: Most lack neighbor awareness section.

Agent Skills: GF(3) Neighborhood Awareness

Install this agent skill to your local

Skill Files