Claude Code Skills Guide: Automating Your Dev Workflow
How Claude Code's Skills evolved into a directory-based standard, with frontmatter controls and a real /code-review workflow example.
Claude Code's Skills feature has evolved from single .md command files into a directory-based format built around a SKILL.md file, now standardized as an open specification via agentskills.io. This update explains how the terminal-based coding agent uses Skills to enforce consistent engineering conventions, covering the new frontmatter fields, argument handling, and dynamic context injection.
The directory structure lets developers keep core instructions lean in SKILL.md while offloading detailed references, scripts, and templates into supporting folders—reducing the token overhead of context that persists throughout a session. New frontmatter fields like disable-model-invocation and user-invocable give explicit control over whether a Skill can be auto-triggered by Claude or only invoked manually, which matters for side-effect-heavy actions like deployments or commits.
Using a real /code-review skill as an example, the guide demonstrates how $ARGUMENTS enables flexible review scopes (staged, branch, fix) and how command output can be pre-executed and injected into the skill's context before the model even sees it. It's a practical reference for engineers who want repeatable, rule-driven code review across a team.