The Claude skills I built for email design (and why they changed how I work)

The most useful thing I've done with AI this year is building a set of reusable skills that make Claude act like a specialist who actually knows email design. There is not one single prompt that can achieve this, but rather continuous training and iterating.

Most people use Claude in the same way: they open a chat, describe what they need, get a response, and move on. That works for one-off tasks, but does not scale to a professional workflow. What scales is a set of configured tools, each built for a specific job, that you can deploy consistently without re-explaining context every time.

In Claude Code and Cowork, those tools are called skills. A skill is a SKILL.md file: a short document with YAML frontmatter that names and describes the skill, followed by structured Markdown that tells Claude exactly what role to take, what task to perform, what constraints to apply, and what output format to return. Once a skill is written, Claude loads it before any task in that domain and follows it every time, without the designer needing to re-establish context.

Here are the skills I built for email design and why each one exists.

The email hierarchy reviewer

This skill instructs Claude to evaluate a described email layout as an experienced email designer focused on mobile rendering. The system prompt defines the role, the task, the constraints (flag anything that competes with the primary CTA, prioritize mobile), and the output format (a short list grouped as critical, recommended, and minor).

Before this skill, getting useful hierarchy feedback from Claude required a paragraph of setup each time. Now I paste in a layout description and get a structured review in under a minute. The things it reliably captures: hero images that share visual weight with the CTA, multi-column product grids where nothing is dominant, and copy blocks that are too long for the available breathing room.

The copy tone auditor

This skill takes email body copy and evaluates it against defined brand voice parameters. The SKILL.md includes examples of approved on-brand copy as reference material, a description of the tone parameters, and an instruction to return flagged passages with a brief explanation of what is off and why.

The value here is consistency across a high-volume email program. Copy drift is real. The tone that felt sharp and intentional in January becomes looser by March. The skill catches that drift at the draft stage before it reaches the subscriber.

The accessibility checker

This skill instructs Claude to evaluate described email design choices against WCAG AA standards, with specific knowledge of how email client rendering affects accessibility. The constraints section lists the common email-specific failure modes: images without meaningful alt text, font sizes below 14px on mobile, color contrast that fails under dark mode inversions, and CTAs that are not distinguishable by color alone.

The output is a checklist-style review: pass, flag, or fail for each evaluated dimension. If you want the full framework, the email accessibility checklist for lifecycle teams covers all of it.

The subject line and preheader evaluator

This skill evaluates a subject line and preheader text as a pair, not as independent elements. The system prompt includes the key evaluation criteria: does the preheader extend the subject line or repeat it, does the combination work at the truncation lengths used by major mobile clients, and does the pair create curiosity or close it too early.

Subject line optimization usually happens in isolation. This skill treats the pair as a unit because that is how subscribers experience them in the inbox. If you need to generate volume first, there's a free AI subject line generator for that. This skill is for evaluating and refining, not generating.

Download the skills

The three skills below are ready-to-use SKILL.md files. Drop them into Claude Code or Cowork and they load automatically whenever you need them.

email-design-qa covers the full pre-development QA pass: visual hierarchy, dark mode, typography fallback, CTA sizing, image alt text, accessibility, and personalization edge cases. It returns a prioritized issue list grouped as critical, recommended, and minor.

email-copy-brand-voice audits email copy against your defined brand voice parameters. Give it a few examples of approved copy, describe your tone rules, and it returns a line-by-line annotation of every passage that drifts off-brand.

email-personalization-spec generates a developer-ready specification for every dynamic content module in an email: merge tags, conditional blocks, product recommendation feeds, empty states, and layout risk flags. It is the spec document that prevents personalization from breaking in production.

How to build your own

A SKILL.md file has two parts. The YAML frontmatter at the top defines the name and description. The Markdown body defines the actual content: the role Claude should take, the specific task it performs, the constraints to follow, any reference examples, and an output checklist.

A few principles that made my skills more reliable over time: give Claude examples of what good and bad output looks like, not just abstract descriptions. Define what Claude should not do as precisely as what it should. Include an output checklist at the end of every skill that Claude runs through before returning a response.

The bigger shift this represents is moving from using AI reactively to using it as a configured part of the workflow. Not "let me ask Claude about this" but "let me run this through the hierarchy reviewer." That shift from reaction to system is where the real leverage lives.

The Figma and Claude integration changed how I move from sketches to code. These skills changed how I evaluate and refine work before it leaves the design phase. Together, they have made the design loop significantly faster without reducing the quality of judgment going into it.