madhvantyagi/SOUL.md

What it solves

It addresses the problem of AI agents having a generic, overly polite, and indistinguishable "default voice." By providing a library of portable persona files, it allows developers to give agents distinct identities, tastes, and convictions that fundamentally change how the agent thinks, pushes back, and makes decisions, rather than just changing its vocabulary.

How it works

The project provides a collection of SOUL.md files. These are Markdown-based personality layers that define an agent's identity, tone, convictions, and behavioral boundaries. These files can be loaded into various AI systems (such as OpenClaw, Hermes, OpenCode, Claude, ChatGPT, or local models) via raw URLs or by pasting the content into system prompts. The framework distinguishes between the agent's character (SOUL.md) and the project-specific rules (AGENTS.md).

Who it’s for

It is designed for developers and AI practitioners who want to create agents with consistent, predictable, and specialized personalities that improve work outcomes—such as an agent that is more disciplined in math or more willing to disagree when a second opinion is needed.

Highlights

• Portable Personas: Works across multiple models and agent tools via simple Markdown files.
• Behavioral Focus: Focuses on changing judgment and prioritization rather than just surface-level speech patterns.
• Structured Framework: Provides a detailed "shape" for creating souls, including sections for uncertainty handling, pushback patterns, and drift checks.
• Empirical Testing: Recommends using the trait-8000 tool and TRAIT benchmark to quantitatively measure if an agent's behavior actually matches the intended personality profile.