What problem does it solve? Writing a well-formed Claude Code skill requires correct frontmatter configuration (name, description, when_to_use, model, effort), a structured SKILL.md body, and triggering boundaries against neighboring skills. Doing this by hand produces skills that fire in the wrong situations, run on the wrong model, or duplicate existing skills. ## Core Features & Use Cases - Guided interview and design: Extracts purpose, triggers, inputs, tools, and naming from the user, then proposes a model × effort pairing from a decision matrix before writing anything. - Convention-driven authoring: Writes SKILL.md from a mandatory template, enforces naming rules (<action>-<object>), the 1,536-character description budget, and documented rationale for every omitted frontmatter field. - Autonomous review loop: Submits the finished skill to an independent review subagent with a clean context, iterates on blocking findings, and escalates to the user only on defined conditions. - Use Case: Ask to create a skill for generating pull request descriptions; the skill interviews you about triggers and tools, drafts SKILL.md per the template, and runs review rounds until the reviewer approves. ## Quick Start Ask the assistant to create a new skill for your repeated workflow, for example: create a skill that drafts release notes from merged pull requests.
