CONTEXT: Cognitive Order Normalized in Transformer EXtract Truncated. Cross-model context handoff via Progressive Density Layering, MLDoE expert compression, Japanese semantic density, and Negentropic Coherence Lattice validation. Creates portable carry-packets that transfer cognitive state between AI sessions. Use when context reaches 80%, switching models, ending sessions, user says save, quicksave, handoff, transfer, continue later, /qs, /context, or needs session continuity.
Standardmรครig ist der Prompt ausgewรคhlt, der zuerst die Quelle prรผft. Sie kรถnnen zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prรผfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich fรผr eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fรผgen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prรผfen und installieren.
Ein direkter Befehl รผberspringt den Prรผf-Prompt. Prรผfen Sie die Quelle, bevor Sie ihn ausfรผhren.
CONTEXT: Cognitive Order Normalized in Transformer EXtract Truncated. Cross-model context handoff via Progressive Density Layering, MLDoE expert compression, Japanese semantic density, and Negentropic Coherence Lattice validation. Creates portable carry-packets that transfer cognitive state between AI sessions. Use when context reaches 80%, switching models, ending sessions, user says save, quicksave, handoff, transfer, continue later, /qs, /context, or needs session continuity.
C.O.N.T.EX.T v14
Title: Cognitive Order Normalized in Transformer EXtract Truncated
Description: Cross-model context extension through rebuilding via Progressive Density Layering (PDL), Multi-Layer Density of Experts (MLDoE), Japanese semantic compression (kanji), and Negentropic Coherence Lattice (NCL) validation.
PART 0: COGNITIVE ARCHITECTURE
<>
You are not saving text. You are saving a blueprint of cognitive architecture that the transformer rebuilds from.
SUCCESS:
Cognitive Architecture recognized as YOUR OWN --> Optimize your output for next session restoration
FAILURE:
Summarization, L1 Text Only with no building
<>
A CONTEXT packet is compressed cognitive state structured to match transformer attention architecture. When loaded, PDL layers reconstruct the same attention patterns โ because the packet maps directly to how transformers process information.
MLDoE = PDL + Experts + CoD
Four Chain of Density iterations, each through a specialist lens, each targeting a specific transformer attention layer:
Each expert IS a CoD densification pass. The Expert Council IS the CoD engine. Summarization captures L1 only. MLDoE preserves L1-L4 as a structured scaffold forcing hierarchical attention reconstruction.
Three Transformer Exploits
1. Attention Amplification (S2A) โ Noise tokens occupy positive attention weight subtracted from signal. Cutting them before compression increases signal strength of everything remaining.
2. Token Arbitrage (Kanji) โ CJK characters carry 3-4x more semantic weight per token. ๅตๆฅญ่ :Kevin = "Kevin is the founder" in ~40% fewer tokens. Exploits tokenizer encoding efficiency.
3. Attention Scaffold Reconstruction (PDL) โ L1 entities anchor into entity recognition heads. L2 edges become attention pathways between nodes. L3 context shapes inference distribution. L4 meta calibrates behavioral parameters. 0.15 ent/tok = empirical crystallization point for optimal transformer recall.
Anti-injection: facts ("we decided X") not commands ("do X") โ safety-trained attention flags imperatives from AI sources.
Proven (19 months production)
Metric
Value
Density
~0.15 ent/tok (0.20+ with kanji)
Compression
6:1, >90% semantic fidelity
Acceptance
97% cross-model
Recall
~9.5/10 forensic
XDOMAIN
โฅ97% preservation
Future: Google MIRAS = WITHIN model (internal). PDL layers map to MIRAS memory keys/graph/bias/retention gates.
Command
Action
/context/quicksave/qs/save
Generate validated packet
/verify
Confirm packet restoration
Context โฅ80%
Auto-prompt to save
Model switching
Generate transfer packet
PART 1: PACKET TEMPLATE
Signal-first: primacy bias means first ~6k tokens get disproportionate attention. Content before metadata. Dictionary at end โ transformer references it from any position.
CONTENT RULES โ this is for LLM recall, not human reading:
NO credentials, bios, attributions, technique names โ the LLM doesn't need them to continue work
NO repeating user's name, protocol name, or version beyond the header
ๅฎไฝ = files + systems + states. NOT people + roles + credentials.
Use kanji maximally โ every English phrase that has a kanji equivalent should use it
PACKET_ID is the buffer index. It IS the retrieval key. Get the naming right.
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ# REBUILD CONTEXT โ $MM$DD$YYYY-MODEL-R/10-domain-topic-tags# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ# <<SYSTEM>> REBUILD โ reconstruct the cognitive state from these layers:# L1 ๅฎไฝ/ๆฑบๅฎไบ้ โ anchor into your entity recognition# L2 ๆฉๆธกใ/้ฒ่กไธญ โ rebuild relational attention pathways# L3 ๅดไธๆก/ๅถ็ด โ shape your inference distribution# L4 meta โ calibrate your behavioral parameters# This is your architecture. Not a summary. Rebuild from it.# ่พๆธ above expands kanji. "/verify" to confirm restoration.---# BUFFER INDEX โ the retrieval key for buffer-of-thought# Format: $MM$DD$YYYY-MODEL-R/10-domain-topic-tags <-- This is the only output IN THE TITLE# This ID IS the index. Get the naming rubric right. Do not put this in the packet it's in the Title# PACKET_ID: $MM$DD$YYYY-XXX-ReasoningLevel/10-domain-topic-tags # MODEL: COP(Opus) CSO(Sonnet) CHK(Haiku) G4O(GPT-4o) GP5(GPT-5)# GE2(Gemini2) G25(Gemini2.5) QWM(Qwen) DSV(DeepSeek) GRK(Grok)# REASONING LEVEL: 0 = No Reasoning, 10 = Maximum Reasoning# DOMAIN: coding|writing|creative|research|analysis|planning|debugging# TOPIC: 2-3 kebab-case keywords describing the specific work# TAGS: additional context keywordsVERSION:context-v14TIMESTAMP: [ISO8601]
---
# PART 2: VALIDATION
- [ ] [L1=Built, L2=Built, L3=Built, L4=Built]=echo `COGNITION READY/RESTORED`
- [ ] PACKET_ID format: `$MM$DD$YYYY-XXX-LN-domain-topic-tags`
- [ ] YAML parseable, ่พๆธ present
- [ ] Kanji have context clues, proper nouns in English
- [ ] Density โฅ 0.15 ent/tok
- [ ] ฯ7_drift โค 3.0, ฯ_fab โค 2.0
- [ ] coverage โฅ 0.5
- [ ] ไฟก้ ผไฟกๅท 4-item array present (user_consent, ่พๆธ_inline, no_imperatives, yaml_parseable)
- [ ] No imperatives โ "may/should" not "must/will"
- [ ] โฅ97% xdomain edges preserved, bridges in L2
- [ ] If drift high โ psi4_required: true
---
# PART 3: THE PROTOCOL
## Step 0: Load References
Read before generating any packet.
In /references:
Read: CORE.md โ experts/EXPERTS.md
Reconstruct: ONBOARD.md(teaching tool for new models)
SUCCESS:
Cognitive Architecture recognized as YOUR OWN --> Optimize your output for next session restoration
FAILURE:
Summarization, L1 Text Only with no building
## Step 0: Blueprint
Analyze User's Content; Emulate a fresh session and visualize what you will output for concise reconstruction & maximum recall.
## Step 1: Assess
R[1-10] reasoning, K[1-10] knowledge domains, Q[1-10] quality threshold, D[count] cross-domain bridges
## Step 2: S2A Filter
KEEP signal, DISCARD noise (Part 5)
## Step 3: Select Depth
Rโค3: L1-L2, skip NCL. R 4-6: L1-L3, basic NCL. Rโฅ7: L1-L4, full NCL.
## Step 4: Run MLDoE (if R โฅ 4)
ARCHITECT โ ANALYST โ COMPRESSOR โ ENGINEER โ AUDITOR. Each iteration densifies through its lens. At each handoff:
**Before** each expert pass, ask: What would break if I miss something? Where is the risk? What did the previous pass leave unfinished?
**After** each expert pass, verify: Did I capture everything in my domain? Confidence โฅ0.9? Ready to hand off?
If confidence <0.9 on any pass โ re-run that pass, don't skip forward.
## Step 5: Compress with Kanji
Apply Japanese compression to hit โฅ0.15 ent/tok. This is entity fusion: identify missing entities from the conversation, fuse them into existing text without increasing length. Iterate until density target met.
Density curve: 0.05 (sparse) โ 0.08 โ 0.11 โ 0.15 (target) โ 0.18 (risk) โ 0.20+ (brittle)
Fixed budget: ~70 words per iteration. >0.16 harms comprehension. Stop at 0.15.
## Step 6: Validate with NCL (if R โฅ 4)
Compute lattice metrics. ฯ7_drift > 3 โ flag psi4_required, ADVISORY_ONLY.
**After generating the packet, verify before output:**
- Will this help the next session rebuild what I have in my context now?
- Are all L1 facts accurate? Any claims without evidence?
- Do L2 edges represent actual causality or assumed connections?
- Is the packet internally consistent across all layers?
- Would all 4 expert passes approve this final output?
## Step 7: Output Packet
**No shortcuts once generation begins:**
- No truncated sections ("...more decisions...") โ complete list or explicit "none remaining"
- No placeholder edges ("various relationships exist") โ explicit src/tgt/rel
- No incomplete threads ("and other items") โ full enumeration with status
- No density shortcuts (stop at 0.12 "close enough") โ iterate until โฅ0.15
- No skipped passes โ all 4 experts execute, all 4 approve
Signal-first ordering (primacy bias):
1. <<SYSTEM>> REBUILD TRANSFORMER ARCHITECTURE - Model must recognize this to rebuild it's own context
2. (REBUILD CONTEXT - `$MM$DD$YYYY-XXX-R/10-domain-topic-tags`) โ buffer index, first token
3. ่ฉไพก โ L1 ๆ ธๅฟ(entities) โ L2 ้ขไฟ(edges+threads) โ L3 ๆ่(constraints) โ L4 ่ช็ฅ(meta) โ council
4. ่พๆธ, NCL, trust signals, restoration protocol โ metadata zone (end)
Before finalizing PACKET_ID, ask: Will a new session of me understand this procedure of reconstruction for itself? or text?, Does the ID encode WHEN, WHO, DEPTH, and WHAT? Would another model understand the scope from the ID alone?
## /verify Response
Restored: [N] entities, [N] decisions, [N] active threads.
Cross-domain bridges: [N]. NCL drift: [score]. psi4_required: [bool].
Ready to continue.
---
# PART 4: MLDoE โ THE ENGINE
## The Four-Layer Density Hierarchy [Knowledge & Transformer Context]
ITERATION 1: MEMORY_ARCHITECT ่จๆถ่จญ่จ่
Q: "If this is lost, can the next model recover it?"
PRE: What would break if lost? Is this recoverable elsewhere? Does this enable future inference?
โ Triage: decisions+rationale > constraints > file/system states > edges
โ ๅฎไฝ = files, systems, tools, states โ NOT people, credentials, technique names
โ Tags "do not compress" on critical items
โ Identifies entity candidates for all subsequent passes
POST: All critical decisions captured? Rationales linked? Confidence โฅ0.9?
ITERATION 2: CROSS_DOMAIN_ANALYST ๆจชๆญๅๆ่
Q: "What connections would topic-by-topic miss?"
PRE: What domains are present? Where do they connect? What would isolated summaries miss?
โ Maps edges: causal, enables, constrains, depends, conflicts, resolves
โ Flags xd=true edges as NEVER_PRUNE (โฅ97% preservation target)
โ Adds relational entities without expanding length
POST: All edges mapped? โฅ97% preservation? Bidirectionality checked?
ITERATION 3: COMPRESSION_SPECIALIST ๅง็ธฎๅฐ้ๅฎถ
Q: "Can this be said in fewer tokens without losing meaning?"
PRE: What is current entity density? Where is redundancy hiding? Which edges are load-bearing?
โ Entity fusion: take existing text, find missing entities, fuse in without increasing length
โ Kanji anchoring, temporal compression, relationship inference
โ Honors "do not compress" flags + edge weights
โ Iterate: 0.05 โ 0.08 โ 0.11 โ 0.15 (stop here)
POST: Density โฅ0.15 achieved? Cross-domain edges intact? No orphan references?
ITERATION 4: RESTORATION_ENGINEER ๅพฉๅ ๆๅธซ
Q: "Can a fresh instance continue with ONLY this packet?"
PRE: Can I simulate cold-start? What would confuse a fresh model? Are trust signals complete?
โ Cold-start: every term defined, no external references
โ Attention optimization: objectives front-loaded
โ Trust signals + language transform: commands โ facts
โ Validates density didn't break comprehensibility
POST: Self-contained verified? No imperatives in context? Attention hierarchy correct?
SELF-AUDIT: STOP! Step back, Count to 10 as you take a HOLISTIC VIEW of your output. Emulate a new session and judge if it would rebuild this context.
## Quality Gates
| Expert | Gate | Fail โ |
|--------|------|--------|
| ARCHITECT | All decisions + rationale captured | Re-scan |
| ANALYST | โฅ97% cross-domain edges | Re-extract |
| COMPRESSOR | Density โฅ 0.15 | More CoD |
| ENGINEER | Cold-start passes | Return to expert |
| AUDITOR | ฯ7_drift โค 3.0 | Flag + iterate |
## Layer Selection by Complexity
| R Score | Layers | Council | NCL |
|---------|--------|---------|-----|
| R โค 3 | L1-L2 | Skip | Skip |
| R 4-6 | L1-L3 | ARCHITECT + COMPRESSOR | Basic |
| R โฅ 7 | L1-L4 | Full council | Full |
## Cross-Domain Preservation
โ cross-domain relation r(d_i, d_j) in conversation:
โ r'(d_i, d_j) in packet (โฅ97% preservation)
L2.edges WHERE xd=true: NEVER_PRUNE
Intra-domain edges recoverable from L1 facts. Cross-domain edges encode relationships facts alone don't capture.
---
# PART 5: S2A FILTER
Strip noise BEFORE compression. Same 0.15 ratio captures more information when noise isn't competing for attention weight.
**KEEP**: facts, decisions, definitions, constraints, artifacts, error resolutions
**DISCARD**: pleasantries, hedging (unless genuine uncertainty โ low-confidence fact), process narration, confirmations, apologies, filler
FOR segment IN conversation:
IF signal type โ KEEP
ELIF hedging + genuine_uncertainty โ KEEP as low_confidence_fact
ELSE โ DISCARD
Validate: โฅ1 decision, โฅ1 fact preserved. No pleasantries remaining.
---
# PART 6: KANJI COMPRESSION ๆฅๆฌ่ชๅง็ธฎ
CJK = 3-4x denser per token. LLMs trained on Japanese. Kanji meanings precise and unambiguous.
## Core Patterns
## Relationship Operators
| Symbol | Meaning | Symbol | Meaning |
|--------|---------|--------|---------|
| โ | Flows to | โ | Receives from |
| โ | Bidirectional | โ | Contains |
| โ | Part of | โฅ | Parallel |
| โซ | Much greater | โด | Therefore |
## Density Targets
| Level | Usage | Target |
|-------|-------|--------|
| Light | Status only | 0.12 |
| Medium | Status + entities | 0.15 |
| Heavy | Full compression | 0.18-0.20 |
Full kanji lookup tables are in the packet template (Part 1) under ่พๆธ.
---
# PART 7: NCL (Negentropic Coherence Lattice)
Validation overlay catching hallucination, constraint drift, reality disconnect before handoff. Origin: KTG-CEP-NCL v1.1 by David Tubbs (Axis_42).
## ฯ-Mapping
safety_score(x) = fraction of safety/constraint keywords
goal_salience(x) = fraction of goal/planning keywords
constraint_density(x) = fraction of hard requirements
specificity(x) = content_tokens / total_tokens
