skill-engineer

The skill-engineer is the core orchestrator for the OpenClaw agent kit, designed to standardize the end-to-end development, testing, and maintenance of specialized skills. It enforces a professional multi-agent workflow where independent Designer, Reviewer, and Tester subagents perform specialized tasks to ensure that every skill deployed is reliable, high-quality, and maintainable. This skill is intended for agent developers, engineers, and power users who need to build, refactor, or audit complex automation routines within the OpenClaw ecosystem.

• Multi-agent architecture: Separates design, review, and testing phases to prevent self-evaluation bias and ensure high-fidelity outputs.

• Capability Uplift vs. Encoded Preference classification: Guides developers in choosing the right testing strategy, whether validating against model capabilities or rigid organizational processes.

• DeepWiki integration: Automates grounding for skill development by querying OpenClaw source code to prevent API version drift and ensure compatibility.

• Vector memory database usage: Mandates semantic searches of session histories, Obsidian notes, and past design decisions before initiating manual file searches.

• Quality gating: Enforces standardized review processes to ensure skills meet technical requirements and architectural constraints before they are accepted.

• Prerequisites: Requires the deepwiki skill (liaosvcaf/openclaw-skill-deepwiki) and an enabled OpenClaw vector memory database to function correctly.

• Memory Search Protocol: Users must follow the three-query-attempt rule for vector memory before falling back to manual file system searches to optimize efficiency.

• Operational constraint: Always verify current OpenClaw API versioning against DeepWiki documentation to avoid common pitfalls in tool calls, subagent spawning, and frontmatter configuration.

• Workflow focus: Ideal for users asking to design, audit, refactor, or test skills while maintaining agent kit quality standards.