skill-creatorlisted

# Skill Creator ## Why this skill exists Agent Skills are the primary mechanism for capturing and reusing complex workflows, and they are consumed by more than one harness: Claude Code and Claude Desktop, Cursor, Codex CLI, Gemini CLI and OpenCode all read `SKILL.md`. A poorly written skill either never triggers (bad description), or triggers but produces inconsistent results (bad body). Most naive attempts produce thin SKILL.md files with vague descriptions and no test coverage — they feel done but fail in practice. This skill provides the full create → test → evaluate → improve loop to get skills to production quality. **Portable first.** Write the Agent Skills core that every harness reads, then add platform extensions where they genuinely help. The reverse order — start from Claude's full field set and hope it degrades — produces skills that carry a vendor's shape everywhere and still fail on the platforms that ignore it. ## Assess where the user is Before doing anything, determine which phase the user is in: | Situation | Start here | |---|---| | "I want to make a skill for X" | [Capture Intent](#1-capture-intent) | | "Turn this conversation into a skill" | Extract from conversation history, then [Write SKILL.md](#3-write-the-skillmd) | | "I already have a draft" | [Run and Evaluate](#running-and-evaluating-test-cases) | | "My skill isn't triggering" | [Description Optimization](#description-optimization) | | "The skill is wrong / produces bad output" | [Improve th