Demonstrates an evolving specification for AI personas in multi-agent systems, suggesting improved integration and safety.
As LLM-driven agents proliferate across consumer and enterprise applications, the lack of a shared specification for AI personas creates fragmented behavior, security gaps, and integration friction. We present Soul Spec, an open specification for declarative AI persona definition with a 5-file structure (SOUL, IDENTITY, AGENTS, TOOLS, USER) optimized for human-readable authoring, machine-verifiable validation, and cross-runtime portability. We trace the evolution of Soul Spec from v0.1 through v0.6 (currently at RFC discussion stage), documenting design decisions driven by real-world deployments. The specification serves as the schema layer for our memory synchronization and identity restoration runtime mechanisms. Concurrent open-source efforts — Karpathy's LLM Wiki and Google Cloud's Scion — validate the design space and clarify Soul Spec's distinctive role as the semantic schema layer that bridges single-agent declarative knowledge management and multi-agent infrastructure orchestration.
No takes yet. Share an insight, caveat, or question.
Tom Jaejoon Lee (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: