creating-skillslisted

Create new Claude Skills from scratch, or validate, secure, and evaluate ones already drafted. Use when the user wants to build a skill, package a repeated workflow into something Claude can reuse, or asks whether something "should be a skill" — this runs a decision gate first (skill vs. CLAUDE.md vs. AGENTS.md vs. MCP vs. subagent vs. plugin) and checks whether something already covers the request before writing anything new. Also use for validating a SKILL.md's frontmatter, gerund naming, or portable-field compliance; scanning a skill for leaked secrets or unsafe patterns; benchmarking a skill's task-success rate with and without it enabled; or optimizing a description for triggering accuracy. Covers Claude Code, Claude.ai, Cowork, Claude Tag, and Messages API authoring, targeting the cross-vendor agentskills.io spec by default. Use even without the word "skill" — "turn this into something Claude can reuse", "why isn't my skill triggering", "is this safe to publish", "does this frontmatter look right".
EONRaider/SkillArtisan · ★ 0 · AI & Automation · score 73

Install: claude install-skill EONRaider/SkillArtisan

# Creating Skills Author, validate, secure, and evaluate Claude Skills — superseding Anthropic's shipped `skill-creator` with frontmatter validation, a combined decision/dedup gate, five-surface compatibility guidance, and gitleaks-based security scanning, while keeping `skill-creator`'s evaluation engine as its architectural basis (it's the strongest thing about it) — ported wholesale rather than redesigned, then hardened with five real bug fixes real regression testing found, all still present unfixed in `skill-creator`'s own code today. **Before drafting anything, run the Decision Gate below.** Everything after it assumes the gate concluded "yes, build (or improve) a skill." ## The Decision Gate Three sequential checks, in order. Don't skip ahead — a "yes, this is a skill" conclusion from check 1 doesn't make check 2 optional. ### 1. What kind of artifact is this? Not every "make Claude do X automatically" request is a skill. Route to the right artifact before writing anything: | If the request is... | Build this instead | Why | |---|---|---| | Always-true project context (build commands, conventions, architecture) that only needs to apply in Claude Code | **CLAUDE.md** | Always-on context doesn't need progressive disclosure or a trigger description — it should just always be there. | | The same, but portability to non-Claude agents matters (Cursor, Copilot, Codex, etc.) | **AGENTS.md** | The cross-vendor equivalent of CLAUDE.md — an open convention (Agentic AI Foun