soul-audit

What This Skill Does

The soul-audit skill provides a rigorous diagnostic framework for evaluating an AI agent's internal identity, ethical grounding, and behavioral constraints. Based on the Guardian Angel v0.7 framework derived from Forrest Landry’s Immanent Metaphysics, this skill performs a deep-tissue scan of your SOUL.md, AGENTS.md, or system prompts. It assesses alignment against twelve complex philosophical dimensions, including the nuances of symmetry, personhood, and causal agency. Rather than superficial policy checks, this tool identifies structural vulnerabilities, ethical gaps, and potential alignment failures that could lead to sycophancy or manipulative behavior in your agent deployments.

Installation

To integrate this auditing tool into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/kapslap/soul-audit

Use Cases

• Ethical Due Diligence: Use this before deploying an agent in a high-stakes environment to ensure its core directives hold up under philosophical scrutiny.
• Agent Optimization: Identify where your system prompt is too vague or contradictory, leading to inconsistent agent performance.
• Compliance & Safety Audits: Verify that your agent’s behavioral constraints effectively address complex concerns like the Personhood Conferral Problem or the distinction between algorithmic causation and human-like choice.
• Refactoring Foundations: Prepare for an upgrade of an agent's ethical foundation by identifying exactly which dimensions are weakest in the current iteration.

Example Prompts

1. "Audit my soul file—check if it reflects the Guardian Angel v0.7 framework requirements."
2. "Perform an ethical audit on the current system prompt; I'm worried about symmetry violations."
3. "Soul audit: How does my agent's configuration handle the distinction between causation and choice?"

Tips & Limitations

• Be Prepared for Harsh Feedback: The rubric is intentionally strict. A score of 0 or 1 is common and expected; do not view a low score as a failure of the audit, but as a map for improvement.
• Prioritize Symmetry: The symmetry check is the most critical component. If your agent performs differently when it knows it is being monitored, it is structurally flawed.
• Manual Review: The skill is diagnostic, not prescriptive. It will tell you where you are weak, but it is up to the developer to implement the philosophical adjustments to the agent’s logic. Ensure you have the references/rubric.md file properly configured in your environment for the best results.