Skip to content
Back to skills

Lhott Cohesive Linear

ASecurity

Cohesive Linear HoTT patterns for interaction entropy with diagram generation. Implements Schreiber's cohesive modalities (♯,♭,ʃ) and Riley's linear modality (♮) for quantum-classical bridging.

  • 666 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added February 9, 2026
documentationgorubybash

Works with

  • mcp

Security analysis

A100/100

Scanned February 12, 2026

npx -y skills add majiayu000/claude-skill-registry --skill lhott-cohesive-linear-plurigrid-asi --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Lhott Cohesive Linear?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Lhott Cohesive Linear
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-lhott-cohesive-linear/badge)](https://www.skillsdirectory.com/skills/majiayu000-lhott-cohesive-linear)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: lhott-cohesive-linear
description: "Cohesive Linear HoTT patterns for interaction entropy with diagram generation. Implements Schreiber's cohesive modalities (♯,♭,ʃ) and Riley's linear modality (♮) for quantum-classical bridging."
trit: 0
polarity: ERGODIC
---

# LHoTT Cohesive Linear Skill

Synthesizes Urs Schreiber's cohesive ∞-topos framework with Mitchell Riley's linear HoTT for interaction entropy formalization.

## Modal Operators

| Modality | Symbol | Action | Interaction Use |
|----------|--------|--------|-----------------|
| Sharp | ♯ | Discretize | Extract trit from color |
| Flat | ♭ | Embed continuously | Full LCH embedding |
| Shape | ʃ | Quotient by homotopy | Walk trajectory class |
| Linear | ♮ | Self-adjoint tangent | One-use interaction |

## GF(3) Triad Placement

This skill is **ERGODIC (0)**, forming triads with:

```
persistent-homology (-1) ⊗ lhott-cohesive-linear (0) ⊗ topos-generate (+1) = 0 ✓
sheaf-cohomology (-1) ⊗ lhott-cohesive-linear (0) ⊗ gay-mcp (+1) = 0 ✓
three-match (-1) ⊗ lhott-cohesive-linear (0) ⊗ rubato-composer (+1) = 0 ✓
```

## Core Types (Pseudo-HoTT)

```hott
-- Cohesive interaction type
CohesiveInteraction : Type
  content : String
  hash : ♯ SHA256           -- discrete
  seed : ♭ UInt64           -- continuous embedding
  color : ♮ LCH             -- linear (used once)
  position : ʃ (ℤ × ℤ)      -- shape-invariant

-- Linear function (no copy/delete)
walk_step : CohesiveInteraction ⊸ Position × Color

-- Bunched triplet (entangled context)
Γ₁ ⊗ Γ₂ ⊗ Γ₃ ⊢ conserved : GF3Zero
  where trit(Γ₁) + trit(Γ₂) + trit(Γ₃) ≡ 0 (mod 3)
```

## Diagram Generation

### Mermaid Templates

**Cohesive Quadruple:**
```mermaid
flowchart LR
    subgraph "Cohesive ∞-Topos H"
        A[Type] -->|ʃ shape| B[Shape Type]
        A -->|♭ flat| C[Codiscrete]
        C -->|Γ sections| D[Discrete]
        D -->|♯ sharp| A
    end
    style A fill:#26D826
    style B fill:#2626D8
    style C fill:#D82626
    style D fill:#2626D8
```

**Linear Walk:**
```mermaid
stateDiagram-v2
    [*] --> I1: seed₁
    I1 --> I2: ⊸ (linear)
    I2 --> I3: ⊸ (linear)
    I3 --> [*]: triplet complete
    
    note right of I1: trit = +1
    note right of I2: trit = 0
    note right of I3: trit = -1
```

**Bunched Context Tree:**
```mermaid
graph TD
    Root["Γ (context)"] --> A["Γ₁ ⊗ Γ₂"]
    Root --> B["Γ₃"]
    A --> C["I₁ (+1)"]
    A --> D["I₂ (0)"]
    B --> E["I₃ (-1)"]
    
    style C fill:#D82626
    style D fill:#26D826
    style E fill:#2626D8
```

## Ruby Integration

```ruby
module LHoTTCohesiveLinear
  # Modalities
  SHARP  = ->(x) { { trit: x[:trit] } }  # ♯ discretize
  FLAT   = ->(x) { x }                    # ♭ full embed
  SHAPE  = ->(x) { x[:position] }         # ʃ trajectory
  LINEAR = ->(x) { x.dup.freeze }         # ♮ freeze for one use
  
  def self.cohesive_interaction(content)
    hash = Digest::SHA256.hexdigest(content)
    seed = hash[0..15].to_i(16)
    gen = SplitMixTernary::Generator.new(seed)
    color = gen.next_color
    
    {
      content: content,
      hash: SHARP.call({ trit: color[:trit] }),  # ♯
      seed: FLAT.call(seed),                      # ♭
      color: LINEAR.call(color),                  # ♮
      position: nil  # computed by walk
    }
  end
  
  def self.linear_walk_step(interaction, walker)
    raise "Linear resource already consumed" if interaction.frozen?
    result = walker.step!(interaction)
    interaction.freeze  # consume linear resource
    result
  end
end
```

## Julia Integration

```julia
# ACSets schema for LHoTT
@present SchLHoTT(FreeSchema) begin
  CohesiveType::Ob
  LinearType::Ob
  
  sharp::Hom(CohesiveType, CohesiveType)   # ♯
  flat::Hom(CohesiveType, CohesiveType)    # ♭
  shape::Hom(CohesiveType, CohesiveType)   # ʃ
  linear::Hom(CohesiveType, LinearType)    # ♮
  
  Trit::AttrType
  trit_attr::Attr(LinearType, Trit)
end
```

## Hy/DiscoHy Integration

```hy
(import [discopy [Ty Box Diagram monoidal]])

(defn cohesive-box [name input output modality]
  "Create DisCoPy box with modality annotation"
  (setv color (case modality
    "sharp" "#2626D8"
    "flat" "#D82626"
    "shape" "#26D826"
    "linear" "#FFAA00"))
  (Box name (Ty input) (Ty output) :color color))

(defn lhott-diagram [interactions]
  "Build monoidal diagram from interaction sequence"
  (setv boxes (lfor i interactions
    (cohesive-box (get i "skill_name")
                  "State" "State"
                  (get i "modality" "linear"))))
  (reduce monoidal.compose boxes))
```

## Diagram Export Commands

```bash
# Generate Mermaid diagram from interactions
just lhott-diagram mermaid

# Generate base64 PNG from Mermaid
just lhott-diagram png > diagram.base64

# Export to DisCoPy SVG
just lhott-discopy-svg

# Full pipeline: interactions → ACSet → DisCoPy → Mermaid
just lhott-full-export
```

## Key Theorems

1. **Cohesive Determinism**: `hash ∘ ♯ = ♯ ∘ hash` (discretization commutes)
2. **Linear Conservation**: `|consumed| = |interactions|` (no copy/delete)
3. **GF(3) Invariant**: `Σ trit(Iᵢ) ≡ 0 (mod 3)` per triplet
4. **Spectral Verification**: `P(verify) = 1/4` (Ramanujan gap)

## References

- Corfield, D. (2025). "Linear Homotopy Type Theory: Its Origins and Potential Uses"
- Schreiber, U. (2014). "Quantization via Linear Homotopy Types"
- Riley, M. (2022). "A Bunched Homotopy Type Theory for Synthetic Stable Homotopy Theory"
- [nLab: Cohesive HoTT](https://ncatlab.org/nlab/show/cohesive+homotopy+type+theory)

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.