Skill Authoring and Evals

Purpose

Make agent instructions reliable without turning the repository into a prompt megapack.

A good skill has one coherent responsibility, a trigger a model can distinguish from neighboring skills, an explicit non-trigger boundary, and a small main body whose branch-specific material is progressively disclosed.

Read Agentit's own docs/SKILL_CURATION.md before adding a permanent skill.

BUILDER workflow

1. Establish the gap

Before writing a new skill, inspect existing Agentit coverage.

Prefer, in order: