Meta-Skills

A meta-skill is a skill that builds other skills. It is the most powerful pattern in the skill engineering toolkit, and it is the pattern that separates operators who use skills from operators who build skill systems. The concept is recursive: instead of manually authoring every skill file, you create a single skill that accepts a workflow description and generates a properly structured skill with front matter, core instructions, and linked file references.

The implementation is straightforward. Your meta-skill's core instructions are a template for skill creation: "Step 1: Ask the user to describe the workflow they want to automate. Step 2: Identify the trigger phrases from their description. Step 3: Decompose the workflow into numbered steps with explicit criteria. Step 4: Identify any templates, examples, or schemas needed as linked files. Step 5: Generate the skill file with YAML front matter, core instructions, and linked file references. Step 6: Save the skill to .claude/commands/ and verify trigger activation."

The meta-skill pattern unlocks a second-order benefit: the reverse meta-prompt. You complete a task manually with Claude — a research report, a code refactor, a data analysis — and then tell Claude to crystallize the process into a skill. "Go through what we just did. Identify the steps, the decision points, and the quality checks. Ignore the noise and false starts. Save the refined process as a skill." Claude extracts the essential workflow from the messy reality of a real session, producing a cleaner skill than you would have authored from scratch.

This is the skill factory loop. Do the task once manually. Crystallize it as a skill. Run the skill on the next occurrence. Observe what the skill gets wrong. Refine the skill. Run it again. After three to five iterations, the skill handles the task better than you could describe it from memory, because it encodes not just the steps but the corrections and edge cases you discovered along the way.