Soulkeeper

What This Skill Does

Soulkeeper is an identity-persistence framework designed to prevent AI agent drift. Agents often lose their core directives, forget available tools, or begin to adopt generic, unhelpful personas during long-running sessions. Soulkeeper solves this by implementing a rules-based audit and reminder system that anchors the agent back to its defined identity found in SOUL.md, TOOLS.md, and AGENTS.md. By utilizing three core scripts—audit.py, drift.py, and remind.py—it ensures that an agent remains consistent with its original programming objectives throughout its entire operational lifecycle.

Installation

Installation is straightforward as the tool utilizes standard Python 3.8+ libraries. First, clone or move the repository into your workspace at /root/.openclaw/workspace/skills/soulkeeper. Once placed, ensure the scripts have execution permissions by running chmod +x audit.py drift.py remind.py. For convenience, it is highly recommended to symlink these scripts to your local PATH, allowing you to trigger them as soul-audit, soul-drift, and soul-remind from any directory within your shell environment.

Use Cases

Soulkeeper is perfect for long-running autonomous agents that need to remain strictly within operational bounds. Developers can use it to perform post-session audits to see if the agent deviated from its instructions. It is also an excellent tool for CI/CD pipelines, where a 'drift threshold' can be set to automatically fail a session if the agent's behavior falls below a certain quality score. By integrating the heartbeat reminder into your session startup, you ensure that every interaction begins with the agent re-aligning itself with its core soul rules.

Example Prompts

1. "Soulkeeper, perform a heartbeat check and refresh my current session rules to ensure I haven't drifted from my core coding directive."
2. "Audit the last five hours of interaction logs from transcript.txt and generate a drift report to identify where I lost my system persona."
3. "Before I begin deploying this code, execute soul-remind with the current context to ensure I am using the approved deployment tools and safety protocols."

Tips & Limitations

To maximize the effectiveness of Soulkeeper, ensure your SOUL.md file is as descriptive and structured as possible. The audit.py script is highly dependent on the quality of your markdown inputs; if your source files are vague, the drift detection will be less effective. Always monitor the drift score; a score above 30 indicates significant behavioral deviation. Finally, while Soulkeeper provides excellent meta-cognitive support, it does not rewrite the agent's internal weights, so it should be viewed as an external governance layer rather than an internal logic change.