ha-skill-creator

Scaffolding a Hope Agent skill involves generating a new directory with a valid SKILL.md file containing correct YAML frontmatter, which the agent reads to recognize and load the custom instructions.

You optimize skill description triggers by running an automated evaluation loop using Claude -p streaming tool-detection, then iteratively refining the description based on should-trigger and should-not-trigger grading outputs to reduce false activations.

YAML frontmatter validation checks the metadata header in a SKILL.md file to ensure the agent parses the skill definition correctly, which is critical for accurate trigger activation when matching user queries.

Yes, you can use automated evaluation and grading to iterate on A/B-style skill description improvements by running trigger audits across should-trigger and should-not-trigger queries, then using the grading outputs to converge on the optimal description.

The best way to audit and clean up existing SKILL.md instructions is to review the YAML frontmatter and description triggers, running an evaluation loop that grades trigger effectiveness to identify and fix misaligned activation behavior.