Spatial Adaptation Artifacts

AI agent platforms are converging toward runtime skill marketplaces — Anthropic Skills, OpenAI Custom Actions, Google Gemini Extensions, Microsoft Copilot Studio. None of them have a structural answer to the questions: which skill applies right now, who certified it, what dependencies must be active, and what happens when a dependency is revoked? This article introduces spatial adaptation artifacts: signed runtime skill loading with admissibility gate as skill router.

Spatial adaptation produces runtime-signed artifacts: each adaptation (skill, policy, model) is signed at runtime by credentialed adaptation authorities. Downstream consumers admit artifacts against the signatures.

Adaptation artifacts undergo sandbox pre-activation certification before operational deployment. The architecture supports declared sandboxing requirements; adaptations failing certification fail admissibility.

Operating units route incoming requests to applicable skills through admissibility evaluation. The admissibility profile, rather than capability matching alone, determines which skill handles each request.

The architecture supports always-active personal layers — user-specific or operator-specific adaptation that operates alongside fleet-default adaptation. Personal layers carry personal credentialing; the architecture composes personal and fleet adaptation through declared composition.

Adaptation deactivation cascades through dependency chains. When an adaptation is deactivated (revoked, expired, superseded), dependent adaptations evaluate continued admissibility; some deactivate structurally as cascade effect.

Adaptations port across underlying models through credentialed translation. An adaptation built for one base model carries declared portability metadata; cross-model translators produce adaptations applicable to alternative base models.

Skill training operates federated across mesh participants. Each participant contributes training observations under credentialed identity; the federated training produces adaptations admitted by all contributing participants.

Adaptations carry composite licensing — intersection of multiple licensing terms from contributing sources. The architecture evaluates the intersection structurally; operations admit only when all relevant licenses admit.

Adaptations distribute peer-to-peer across the mesh rather than from central distribution authority. Each unit can contribute adaptation distribution to other units; credentialed distribution chains support the decentralized distribution.

Defense field operations require runtime adaptation that respects mission-specific operational profiles. The spatial-adaptation primitive supports field-deployed adaptation under mission-specific governance without compromising operational integrity.

Industrial-robotics runtime adaptation requires safety-aware adaptation that supports continuous improvement while maintaining safety-critical operational integrity. The spatial-adaptation primitive supports industrial-robotics adaptive operation.

Medical-device runtime adaptation (especially for AI-augmented medical devices) requires regulatory-aware adaptation that supports continuous improvement while maintaining FDA-grade safety governance. The spatial-adaptation primitive supports medical-device adaptive operation.

Cybersecurity rapid-update operations (zero-day vulnerability response, ransomware-containment update, supply-chain-attack response) face structural deployment challenges. Architectural spatial-adaptation supports cybersecurity rapid-update operations.

LLM adaptation under emerging EU AI Act, U.S. AI Executive Order, and similar regulatory frameworks requires structural adaptation governance beyond platform-internal handling. Architectural spatial-adaptation supports regulatory-aware LLM adaptation.

IEC 62304 establishes the international standard for medical-device software lifecycle. Spatial-adaptation provides the architectural substrate for emerging continuously-improving medical-AI software.

NIST AI Risk Management Framework (AI RMF 1.0) establishes the dominant U.S. AI risk management framework with substantial adoption across federal agencies and commercial operators. Spatial-adaptation substrate provides architectural foundation.

UN ECE R156 establishes Software Update Management System (SUMS) requirements for vehicle-OEMs across UNECE-1958 contracting parties. The structural requirements integrate directly with spatial-adaptation's runtime-signed adaptation primitive.

Anthropic Skills provides AI-agent capability framework. The architectural element above Anthropic Skills — runtime-signed adaptation with cross-authority composition supporting multi-jurisdiction operations — is what spatial-adaptation primitive provides.

OpenAI offers fine-tuning of GPT models for customer-specific adaptation. The architectural element above OpenAI fine-tuning — runtime-signed adaptation with composite admissibility and cross-jurisdiction governance — is what spatial-adaptation primitive provides.

Tesla FSD over-the-air updates operate at significant fleet scale. The architectural element above Tesla updates — runtime-signed adaptation artifacts with sandbox pre-activation and cascade-deactivation — is what spatial-adaptation primitive provides.

AWS Bedrock operates managed-LLM platform. Architectural element — adaptation governance substrate — is what spatial-adaptation provides.

Databricks operates major commercial data-and-AI platform. Architectural element — adaptation governance — is what spatial-adaptation provides.

Google Vertex AI operates managed-ML platform. Architectural element — adaptation governance — is what spatial-adaptation provides.

Hugging Face Hub operates major commercial AI-model sharing platform. Architectural element — adaptation governance — is what spatial-adaptation provides.

Anthropic operates Claude foundation model with Model Context Protocol (MCP) for agent capability composition and Claude Skills for runtime adaptation. Architectural element above MCP — credentialed adaptation governance with composite admissibility — is what spatial-adaptation provides for emerging cross-jurisdiction agent operations.