trace-to-skill-converter

To convert a workflow trace into a reusable skill definition, run the trace-to-skill converter on a markdown trace file. It deterministically extracts structured data and delegates semantic naming to an LLM to generate a draft SKILL.md.

Deterministic extraction parses traces, including process trees and artifacts, into structured JSON without LLM involvement. This ensures repeatable, structured data generation before the LLM derives semantic names and instructions.

The best way to automate skill generation from project traces is using a phased workflow that separates deterministic extraction, LLM synthesis, and architecture evaluation with governance gates, ensuring repeatable and testable skill outputs.

Yes, a valid SKILL.md frontmatter is required to generate a draft skill from a trace. The converter uses this frontmatter to output a JSON contract for phase 2 synthesis and subsequent architecture evaluation.

Yes, you can use workflow traces with optional process trees, artifacts, and phased sections to create testable skills. The converter applies deterministic extraction to diverse trace structures to enable repeatable skill generation across projects.

The conversion process delegates architecture evaluation to a sub-agent during the assembly and evaluation phase. This ensures the generated skill meets governance gates and quality standards before finalizing the SKILL.md for team reuse.