skill-auto-improver

You can automate SKILL.md improvements by running a phase-driven loop that applies deterministic fixes and frontmatter normalization. This process retrofits existing local files to meet the skill-creator Gate 1 standard and the asm-eval 85/8 quality floor.

The asm-eval 85/8 floor is a required quality threshold that SKILL.md files must pass during automated improvement. It represents a scoring baseline evaluated by the asm tool to ensure skills meet catalog readiness before publishing.

Yes, normalizing frontmatter and running the improvement loop requires Python 3, the asm tool on PATH, and the skill-creator validator. These dependencies are necessary to execute the deterministic fixes and validation checks.

Migrating top-level author and version fields involves moving them into the metadata block to satisfy quick_validate.py. The auto-improver handles this automatically during the Gate 1 phase to normalize the overall frontmatter structure.

When validation fails, the automation applies targeted per-category edits and version bumps iteratively. If a blocker prevents passing both gates, the loop stops and emits a final report describing the unresolved issues and changes.

Yes, the process generates .asm-improver artifacts and a final report. These outputs document the applied deterministic changes, per-iteration version bumps, and the overall improvement status for the retrofitted SKILL.md.