skill-authoring-lifecycle

Create a kebab-case directory under skills/ matching the frontmatter name, copy the SKILL.md template, and write a description covering capability and trigger phrases. Then register the skill id in skills.sh.json and add a row to the README skill table.

Check that name is 1-64 lowercase alphanumeric characters with hyphens, description is 1-1024 chars stating capability and activation triggers, and metadata includes version and author. Run pnpm run validate to confirm no warnings.

The description must state both what the skill does and when to activate it, using trigger phrases users actually say. Activation-critical routing belongs in the description, not only in a body section.

Yes, but the skill must run from a copied directory without relying on repo-relative paths outside its package. Agent-specific assumptions should be isolated to optional metadata like agents/openai.yaml, not the core workflow.

T1 behavior-critical skills require references/evals.md plus at least two evals/cases/*.yaml files covering should-trigger and should-not-trigger prompts. Other skills may include evals optionally.