skill-forge

Gate-driven skill development enforces quality assurance by applying a unified 6-step workflow with automated validation, CSO scoring, and subagent testing before skill deployment. This process reduces drift and ensures consistent quality across projects.

The best way to package a Claude skill is using a gate-driven workflow that runs lint validation, CSO scoring, and subagent testing. You execute init, lint, score, test, and package scripts sequentially to ensure the skill is ready for deployment.

CSO scoring quantifies skill description quality to trigger discovery reliability. By running the score_description.py script, you evaluate and score the SKILL.md file, ensuring the description meets automated validation standards for better searchability.

Yes, PyYAML is a required dependency for the gate-driven skill development lifecycle. You need the PyYAML package installed in your environment to run the provided scripts for initializing, linting, scoring, testing, and packaging skills.

To test a Claude skill with a subagent, run the test_skill.py script targeting your skill directory. This subagent testing step validates the skill's behavior and ensures repeatability before you proceed to the final packaging and publishing phase.

Use a formal 6-step workflow for skill development when you need to reduce drift and ensure consistent quality across team projects. The automated validation and subagent testing prevent errors that occur when manually editing and deploying skill files without structural checks.