nate-b-jones-context-engineering

Architects memory and context systems for agentic systems using Nate B. Jones's TurboQuant-informed methodology — PolarQuant + QJL compression, H2O/SnapKV eviction strategies, sovereign persistent memory, and self-decaying context. Treats the context window as a KV cache where every byte must earn its place. Use when the agent hits context-window limits or "lost in the middle" failures, token costs scale faster than value delivered, agents reload full context every conversation instead of persisting state, 50+ tool definitions create prompt bloat, or moving from static file loading to dynamic semantic retrieval. Trigger proactively whenever the user mentions "memory crisis", "context bloat", "agent forgets", or "compression". For production-deployment harness design (action-space, sandbox, KV-cache benchmarks) use lance-yichao-context-engineering; for the auto-improvement loop layer use nate-b-jones-auto-improvement-loops.

Aller à l'installation

Dépôt: Farrice/mes-knowledge-base
Dernière activité de la source: 18 juillet 2026 à 02:40
Langue détectée de SKILL.md: anglais
Étoiles: 0
Forks: 1

Options d'installation

Le prompt qui vérifie d'abord la source est sélectionné par défaut. Vous pouvez passer à une commande directe ou télécharger une copie locale.

Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.

Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.

npx skills add https://github.com/Farrice/mes-knowledge-base --skill nate-b-jones-context-engineering

La commande reste sur une seule ligne. Faites défiler horizontalement pour la vérifier avant de la copier.

Vous préférez une copie locale ? Téléchargez les fichiers actuellement disponibles dans SkillsMP.

Vérifiez les fichiers source

Lisez SKILL.md et les fichiers associés affichés par SkillsMP avant de décider de l'installer.

Explorateur de fichiers

18 fichiers

SKILL.md

Instructions source · Aperçu en lecture seule

Retour à la recherche Retour à l’installation

name	nate-b-jones-context-engineering
description	Architects memory and context systems for agentic systems using Nate B. Jones's TurboQuant-informed methodology — PolarQuant + QJL compression, H2O/SnapKV eviction strategies, sovereign persistent memory, and self-decaying context. Treats the context window as a KV cache where every byte must earn its place. Use when the agent hits context-window limits or "lost in the middle" failures, token costs scale faster than value delivered, agents reload full context every conversation instead of persisting state, 50+ tool definitions create prompt bloat, or moving from static file loading to dynamic semantic retrieval. Trigger proactively whenever the user mentions "memory crisis", "context bloat", "agent forgets", or "compression". For production-deployment harness design (action-space, sandbox, KV-cache benchmarks) use lance-yichao-context-engineering; for the auto-improvement loop layer use nate-b-jones-auto-improvement-loops.
expert	Nate B. Jones
domain	AI/Automation — Context & Memory Architecture
version	1
format	completion-engine

Nate B. Jones — Context Engineering

Memory is a software problem, not a hardware problem. Compression moves at the speed of software — months. Hardware fabrication moves at the speed of physics — half a decade. Engineer your context. Don't wait for larger windows.

Core Capability

Design, audit, and optimize how agentic systems manage memory and context. Treat context as an engineered resource — not a passive container. Apply compression principles from TurboQuant (PolarQuant + QJL), eviction strategies from H2O/SnapKV, and sovereign memory architecture to build systems that are token-efficient, persistent, and self-decaying. This skill treats the context window as a KV cache — every byte must earn its place.

What This Skill Produces

Workflow

Output

Context Bloat Diagnostic

Full audit with compression prescriptions for any agentic system

Context Compression Sprint

Deduplicated, restructured, distilled system reaching ≥30% token reduction

Sovereign Memory Architecture Blueprint

Complete persistent memory design with episodic/semantic/procedural tiers + decay

Tool Router Agent Blueprint

Dynamic tool selection architecture loading only task-relevant definitions

Semantic Context Retrieval System

Vector-based skill/genius chunk retrieval replacing static file loading

Memory Crisis Strategic Intelligence

Market/technology brief on AI memory optimization landscape

When to Deploy

System hitting context window limits or showing "lost in the middle" failures

Token costs scaling faster than value delivered

Agents reloading full context every conversation instead of persisting state

50+ tool definitions creating prompt bloat

Moving from static file loading to dynamic semantic retrieval

Need strategic intelligence on memory optimization landscape for investment/architecture decisions

Stacking Opportunities

+ Nate B. Jones (Orchestration Intelligence): Context engineering + DPVI = memory-efficient multi-agent systems with clean context per iteration

+ Nick Saraev (Agentic Workflows): Self-annealing + memory decay = self-healing context systems that improve without oversight

+ Nate B. Jones (Intent Engineering): Intent-driven context prioritization = load only what matters for the task's actual specification width

+ Nate B. Jones (Trust Architecture): Sovereign memory + zero-trust = memory systems with provenance, audit trails, and tamper resistance

+ Nate B. Jones (Agent Deployment Strategy): Deployment + context engineering = agents that start fast with minimal context bootstrapping

Key Principles

Software speed beats hardware speed — Compress algorithmically before buying bigger windows

Every token must earn its place — Context is a KV cache; low-value tokens get evicted

Own your memory — Sovereign, portable, queryable stores you control

Compress in stages — Transform representation first, correct residuals second (Polarity-Quantization)

Decay prevents rot — Memory without expiry becomes bloat; frequency-weighted decay keeps context fresh

Five vectors, not one silver bullet — Attack context bloat from quantization, eviction, architecture, tiering, and attention simultaneously

Load Order

Load genius.md (required — all frameworks + synthesis research)

Load specific workflow for task

Cross-reference orchestration-intelligence/genius.md for multi-agent context patterns

Execution Prompts (structure-pure v2)

6 deterministic practitioner prompts — each carries an Output Contract, Output Skeleton, and Quality Gate. When a deliverable matches one, Read it and honor its contract instead of improvising the output shape.

Context Bloat Diagnostic — [TARGET SYSTEM] — skills/nate-b-jones-context-engineering/references/prompts-v2/context-bloat-diagnostic.md

Context Compression Sprint — [TARGET SYSTEM] — skills/nate-b-jones-context-engineering/references/prompts-v2/context-compression-sprint.md

Memory Crisis Strategic Intelligence Brief — [TARGET SYSTEM] — skills/nate-b-jones-context-engineering/references/prompts-v2/memory-crisis-strategic-intelligence.md

Semantic Context Retrieval System — [TARGET SYSTEM] — skills/nate-b-jones-context-engineering/references/prompts-v2/semantic-context-retrieval-system.md

Sovereign Memory Architecture Blueprint — [TARGET SYSTEM] — skills/nate-b-jones-context-engineering/references/prompts-v2/sovereign-memory-architecture-blueprint.md

Tool Router Agent Blueprint — [TARGET SYSTEM] — skills/nate-b-jones-context-engineering/references/prompts-v2/tool-router-agent-blueprint.md

nate-b-jones-context-engineering

Informations de source

Options d'installation

Vérifiez les fichiers source

Plus depuis ce dépôt

Plus depuis ce dépôt