self-improving-to-expertpack

What This Skill Does

The self-improving-to-expertpack skill acts as a bridge between the iterative development cycle of a Self-Improving Agent and the standardized, high-performance architecture of an ExpertPack. As an agent evolves, it accumulates a repository of tribal knowledge—best practices, error logs, and feature requests—often trapped in raw files like LEARNINGS.md or ERRORS.md. This skill automates the migration of these disparate notes into a clean, schema-compliant ExpertPack (v2.3). It performs intelligent parsing, deduplication, and restructuring, moving content into specialized directories like /mind, /facts, and /operational. Crucially, it secures your data by stripping secrets like API keys or tokens during the transformation process, ensuring your exported knowledge base remains safe for sharing or deployment.

Installation

Installation is handled via the OpenClaw platform. Use the standard install command within your terminal:

clawhub install openclaw/skills/skills/brianhearn/self-improving-to-expertpack

Ensure your local environment has the required Python dependencies installed, as the script leverages Python 3 for its parsing and normalization logic. Once installed, the tool is available as a command-line utility within your workspace environment.

Use Cases

This skill is designed for developers and AI operators who have reached a saturation point with their current agent's history and want to move toward a more scalable, version-controlled architecture. Common use cases include:

• Migrating from a prototyping phase to a production-grade ExpertPack deployment.
• Backing up accumulated agent knowledge before switching between different LLM backends or agent platforms.
• Consolidating scattered documentation into a centralized, searchable, and machine-readable format.
• Preparing an agent's logic for integration into a team-wide repository where structured meta-data and context tiers are required.

Example Prompts

• "self-improving to expertpack: convert my current workspace learnings into a new pack named 'production-knowledge-base' in ~/expertpacks/"
• "export learnings from the agent and initialize an expertpack with the type 'process' to ensure all my procedural notes are categorized correctly."
• "migrate self-improving files: please run the conversion script on my current .learnings/ directory and output the manifest to ./expertpack-v1"

Tips & Limitations

• Verification: Always review the generated overview.md after conversion to ensure that the priority mapping aligns with your project goals.
• Secrets: While the skill attempts to strip credentials, verify sensitive files manually if you have included highly non-standard patterns for tokens.
• Chunking: Don't forget to run the chunk.py post-migration; ExpertPacks rely on these chunks for efficient retrieval performance.
• Constraint: This skill specifically requires a structured .learnings/ folder. If your agent uses custom log locations, use the --learnings flag to point the script to the correct directory.