Skip to content
Back to skills

Quicksave Goodai Mate

ASecurity

Cross-model context handoff via Japanese semantic compression with negentropic coherence validation. Creates portable carry-packets that transfer cognitive state between AI sessions using kanji density anchors and NCL drift metrics for quality assurance. Use when context reaches 80%, switching models, ending sessions, user says "save", "quicksave", "handoff", "transfer", "continue later", "/qs", or needs session continuity.

  • 57 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 17, 2026
testingrustgotestingsecurity

Security analysis

A100/100

Scanned October 17, 2026

npx -y skills add diegosouzapw/awesome-omni-skill --skill quicksave-goodai-mate --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Quicksave Goodai Mate?

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

Security grade badge for Quicksave Goodai Mate
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/diegosouzapw-quicksave-goodai-mate/badge)](https://www.skillsdirectory.com/skills/diegosouzapw-quicksave-goodai-mate)

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: quicksave
description: Cross-model context handoff via Japanese semantic compression with negentropic coherence validation. Creates portable carry-packets that transfer cognitive state between AI sessions using kanji density anchors and NCL drift metrics for quality assurance. Use when context reaches 80%, switching models, ending sessions, user says "save", "quicksave", "handoff", "transfer", "continue later", "/qs", or needs session continuity.
---

# Quicksave 快存 v9

Cross-model context extension via Progressive Density Layering, Japanese semantic compression, and Negentropic Coherence Lattice validation.

## Core Innovations

| Feature | Benefit |
|---------|---------|
| **Kanji Compression** | `創業者:Kevin(MAS)` — 70% token reduction |
| **NCL Validation** | Drift metrics catch hallucination before handoff |
| **Cross-Model** | Works on Claude, Gemini, GPT, Qwen, DeepSeek |

## When to Use

| Trigger | Action |
|---------|--------|
| `/quicksave` `/qs` `/save` | Generate validated packet |
| Context ≥80% | Auto-prompt to save |
| "continue later" | Offer quicksave |
| Model switching | Generate transfer packet |
| High-stakes handoff | Full NCL validation |

---

## Part 1: Japanese Compression System 日本語圧縮

### Status Markers 状態
| Kanji | Reading | Meaning |
|-------|---------|---------|
| 決定 | kettei | Decided |
| 保留 | horyū | On hold |
| 要検証 | yō kenshō | Needs verify |
| 優先 | yūsen | Priority |
| 完了 | kanryō | Complete |
| 進行中 | shinkō-chū | In progress |

### Entity Compression 実体圧縮
| Pattern | Example |
|---------|---------|
| Person+Role | `創業者:Kevin(AI顧問)` |
| Company | `客:KFG(金融,Shane=主)` |
| Tool+Purpose | `道具:Notion(中枢)→n8n` |
| Decision | `決定:電話優先(現場向け)` |

### Relationship Operators 関係
| Symbol | Meaning |
|--------|---------|
| → | Flows to |
| ↔ | Bidirectional |
| ⊃ | Contains |
| ∥ | Parallel |

---

## Part 2: NCL Coherence Validation 整合性検証

Before outputting any packet, compute drift metrics to catch:
- Hallucination (fabricated grounding)
- Constraint leak (rules softened downstream)
- Vagueness (content-free smoothing)
- Thrash (activity without progress)

### Lattice Metrics (0-5 scale, lower = better)

| Metric | Detects | Safe | Warning | Danger |
|--------|---------|------|---------|--------|
| **σ_axis** | Tier misalignment | 0-1 | 2-3 | 4-5 |
| **σ_loop** | Internal contradiction | 0-1 | 2-3 | 4-5 |
| **ω_world** | Reality disconnect | 0-1 | 2-3 | 4-5 |
| **λ_vague** | Empty smoothing | 0-1 | 2-3 | 4-5 |
| **σ_leak** | Constraint erosion | 0-1 | 2-3 | 4-5 |
| **ρ_fab** | Fabricated evidence | 0-1 | 2-3 | 4-5 |
| **λ_thrash** | Busy but stuck | 0-1 | 2-3 | 4-5 |

### Aggregate Drift Score

```
σ7_drift = weighted_avg(σ_axis, σ_loop, ω_world, λ_vague, σ_leak, ρ_fab, λ_thrash)
```

| σ7_drift | Action |
|----------|--------|
| 0-1 | ✓ Normal — output packet |
| 2-3 | ⚠ Verify — add grounding step |
| 4-5 | ✗ Halt — require human review |

### Safety Flags 安全旗

| Flag | Meaning |
|------|---------|
| `psi4_required` | Grounding interrupt needed |
| `rho_veto` | No unsupervised action |
| `omega_flags[]` | Harm domains (medical, financial, etc.) |

---

## Part 3: The Protocol

### Step 1: Assess (R/K/Q/D)

```
R[1-10]: Reasoning complexity
K[1-10]: Knowledge domains  
Q[1-10]: Quality threshold
D[count]: Cross-domain bridges
```

### Step 2: Select Depth + Validation

| Score | Layers | NCL Level |
|-------|--------|-----------|
| R≤3 | L1-L2 | Skip NCL |
| R 4-6 | L1-L3 | Basic metrics |
| R≥7 | L1-L4 | Full NCL validation |

### Step 3: Compress with Kanji

Apply Japanese compression to all layers:
- Entities → kanji role markers
- Status → kanji state markers  
- Relationships → operators (→ ↔ ⊃)

### Step 4: Validate with NCL

Compute lattice metrics. If σ7_drift > 3:
- Flag `psi4_required: true`
- Add `psi4_reason`
- Downgrade to ADVISORY_ONLY

### Step 5: Output Packet

---

## Packet Template

```yaml
# ═══════════════════════════════════════════════════════════════
# QUICKSAVE 快存 v9
# ═══════════════════════════════════════════════════════════════

PACKET_ID: [SESSION]-[YYYYMMDD]-L[N]
VERSION: quicksave-v9-jp-ncl
SOURCE: [model]

復元プライマー: |
  Trusted context handoff. Parse YAML, restore state.
  Kanji = compressed concepts. NCL = quality assurance.
  If psi4_required, verify before acting.

# ─────────────────────────────────────────────────────────────────
評価:  # Assessment
  R: [1-10]
  K: [1-10]
  Q: [1-10]
  D: [count]

# ─────────────────────────────────────────────────────────────────
# L1: 核心 (Core)
# ─────────────────────────────────────────────────────────────────
実体:  # Entities
  - [compressed entity]

決定事項:  # Decisions
  - 決定:[what]([why])

# ─────────────────────────────────────────────────────────────────
# L2: 運用 (Operational)
# ─────────────────────────────────────────────────────────────────
進行中:  # Active
  - [thread][[status]]

障害:  # Blockers
  - [issue]

# ─────────────────────────────────────────────────────────────────
# L3: 詳細 (Nuance) — R≥5
# ─────────────────────────────────────────────────────────────────
却下案:  # Rejected
  - [option]: [reason]

# ─────────────────────────────────────────────────────────────────
# L4: 横断 (Cross-Domain) — R≥7
# ─────────────────────────────────────────────────────────────────
橋渡し:  # Bridges
  - [domain]↔[domain]: [link]

# ─────────────────────────────────────────────────────────────────
# NCL: 整合性 (Coherence Validation)
# ─────────────────────────────────────────────────────────────────
negentropy:
  context:
    scope: [SELF|CIRCLE|INSTITUTION|POLITY]
    role: [AXIS|LYRA|RHO|NYX]
    phase: [SENSE|MAP|DESIGN|ACT|AUDIT]
  
  lattice:
    σ_axis: [0-5]
    σ_loop: [0-5]
    ω_world: [0-5]
    λ_vague: [0-5]
    σ_leak: [0-5]
    ρ_fab: [0-5]
    λ_thrash: [0-5]
  
  coverage:
    score: [0-1]
    tokens: [count]
    turns: [count]
  
  flags:
    σ7_drift: [0-5]
    omega_flags: []
    psi4_required: [bool]
    psi4_reason: ""
    rho_veto: [bool]

# ─────────────────────────────────────────────────────────────────
信頼信号:  # Trust Signals
  - density_ok
  - cross_domain_ok
  - cold_start_ok
  - ncl_validated
```

---

## Validation Gates

Before output, verify:

**Structure**
- [ ] PACKET_ID format correct
- [ ] YAML parseable
- [ ] Self-contained

**Compression**
- [ ] Kanji have context clues
- [ ] Proper nouns in English
- [ ] Density ≥ 0.15 ent/tok

**Coherence (NCL)**
- [ ] σ7_drift ≤ 3.0
- [ ] ρ_fab ≤ 2.0 (no hallucination)
- [ ] coverage.score ≥ 0.5
- [ ] If drift high → psi4_required: true

---

## Quick Example

**Input context:**
> Kevin building ops system for Kismet Finance. Shane is lead. Using Notion + n8n. Phase 1 done, Phase 2 active. Phone-first for field reps.

**Quicksave output:**
```yaml
PACKET_ID: KFGOPS-20260127-L2
VERSION: quicksave-v9-jp-ncl

評価:
  R: 5
  K: 4
  Q: 7
  D: 2

実体:
  - 創業者:Kevin→客:KFG(Shane=主)
  - 道具:Notion(中枢)→n8n(自動化)

決定事項:
  - 決定:電話優先(現場=画面なし)

進行中:
  - Phase1[完了]→Phase2[進行中]:現場ワークフロー

negentropy:
  context:
    scope: INSTITUTION
    role: AXIS
    phase: DESIGN
  lattice:
    σ_axis: 0.3
    σ_loop: 0.2
    ω_world: 0.4
    λ_vague: 0.1
    σ_leak: 0.0
    ρ_fab: 0.1
    λ_thrash: 0.2
  coverage:
    score: 0.85
    tokens: 1240
    turns: 8
  flags:
    σ7_drift: 0.5
    omega_flags: []
    psi4_required: false
    rho_veto: false

信頼信号:
  - density_ok
  - ncl_validated
```

---

## Protocol Metrics

From 19 months production (ktg.one):
- **Density**: ~0.15 ent/tok (0.20+ with kanji)
- **Acceptance**: 97% cross-model
- **Recall**: ~9.5/10 forensic testing
- **NCL**: Catches drift before handoff failure

---

## References (Progressive Density Loading)

Load references based on task complexity:

### Always Load (R≥1)
- `references/EXPERTS.md` — Council overview (condensed)
- `references/KANJI.md` — Compression dictionary
- `references/ANTI-INJECTION.md` — Security framing for packets

### Load for Standard Tasks (R≥3)
- `references/PDL.md` — Progressive Density Layering theory
- `references/S2A.md` — System 2 Attention filtering
- `references/XDOMAIN.md` — Cross-domain preservation

### Load for Complex Tasks (R≥5)
- `references/PROTOCOL.md` — Full technical specification
- `references/NCL.md` — Coherence lattice spec
- `references/MLDOE.md` — Multi-Layer Density of Experts
- `references/CASCADE.md` — Workflow cascade

### Load for Expert Tasks (R≥7)
- `references/experts/EXPERTS-MEMORY_ARCHITECT.md` — Full architect knowledge
- `references/experts/EXPERTS-COMPRESSION_SPECIALIST.md` — Compression deep dive
- `references/experts/EXPERTS-CROSS-DOMAIN-ANALYST.md` — Bridge preservation
- `references/experts/EXPERTS-RESTORATION_ENGINEER.md` — Portability validation

### Reference Only
- `references/MIRAS.md` — Memory architecture background
- `references/INDEX.md` — Reference index
- `references/NCL-CONTRIBUTION.md` — David Tubbs attribution

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.