#73 | Claude Skills: Build the right ones

TL;DR: Claude Skills let you encode a recurring workflow once and trigger it in any session, indefinitely. This edition covers the eligibility rule that separates what belongs as a Skill from what should stay as a saved prompt.

👋 Hello,

If you are like me and use AI regularly, you probably have a prompt document somewhere. For accessibility reasons, I usually store mine in Notes, Notion, TextBlaze, and Cursor. I’m sure there are better and more effective options.

Well, over time, that library grows. Finding the right prompt mid-session carries its own friction, too. At some point, the retrieval takes as long as writing a new one.

I’m sure you’ve heard about the fuss about Claude Skills and Claude Cowork. If not (don’t worry), a skill is simply a packaged instruction set. It is triggered by a slash command that requires no session setup.

Once installed, it works in any future chat, indefinitely. Developers, enthusiasts, and even small-business users have been building with them for several months now. Again, no rush here; that is probably still only 0.0001% of the population.

Well, today I’d like to share a resource I found earlier this week. A YouTube creator named Alex (Grow with Alex) published a tutorial in May showing six Skills running his entire social media workflow.

This edition covers that system and the decision rule at its center. By the end, you will know which of your saved prompts belongs as a Skill.

Questions for you: Do you want to read more about Claude Skills and Cowork in future editions? Reply to this email, and say: YES.

How Skills work in practice

Ok, so Claude Skills have been available since October 2025. AI practitioners have been documenting what they make possible — newsletter writers, coders, content operators, and legal teams.

​Alex’s video​ is among the more complete workflow builds shared publicly. ​The Neuron Daily covered it​ three days later, on May 14. That is where I found it.

He built six Skills to cover his complete social media content production cycle. Each has a slash command.

Type it in any chat, and Claude loads the full instructions for that task immediately, with no context-setting required.

• /hook — Hook generator. Generates content hooks in the creator’s voice, with style rules and banned phrases baked in.
• /b_image — Image generator. On-brand b-roll images via ​Higgsfield MCP​; branding rules embedded.
• Video generator. Extends image sequences into cinematic video clips. Slash command not stated in the video — see the vault below for the full skill file.
• Thumbnail generator. Analyzes top-performing thumbnails, builds a style profile, and generates new ones to match.
• Face lock. Stacks on top of the thumbnail skill. Injects identity-preservation language to prevent facial distortion during generation.
• /IG_captions — Caption funnel. Platform-specific captions with embedded ManyChat automation language.

Alex’s full skill files — including the exact trigger commands for the video generator, thumbnail generator, and face lock — are available free in ​his vault​.

The skills also stack. His thumbnail generator and face lock run together — two focused tools combining to handle something neither could solve alone.

Together, they produce something a single larger Skill would handle less reliably.

Every prompt, every style rule, every platform convention lives in his skills folder. Available in any chat, indefinitely, without reconfiguration.

Ok, but how does the loading work?

At session start, Claude reads only the name and description of each installed Skill — around 100 tokens per Skill.

Full instructions load only when a Skill is triggered. Twenty Skills installed means roughly 2,000 tokens at startup.

Loading the same instructions manually incurs significantly higher costs before any task begins. In practical terms, installing more Skills keeps sessions lean.

Alex states the decision rule directly in the video. A task earns a Skill when it is recurring, has a fixed structure, and produces consistent output. All three required. Pass two out of three, and it stays as a saved prompt.

The eligibility rule
A task earns a Skill when it is recurring, structured, and produces consistent output. All three required. Two out of three: keep it as a saved prompt.

Of the three, consistent output is where most prompts fail to qualify.

Build an unreliable task as a Skill, and it produces that unreliability more consistently. The filter exists to protect the compounding reliability that makes Skills worthwhile.

A warning on token windows and credit usage

I have a Claude skill to monitor, read, and filter AI newsletters for topics in my niche and interests. I need to because otherwise I might only be able to consume 3-4 newsletters rather than 100+.

Each scheduled scan consumed roughly 80% of the current session budget of my Claude Pro Plan. That is the actual number.

One large-context automated task moved the counter most of the way to zero before any real work had begun.

By the time the newsletter reading finished, the session sat at about 20% remaining. A single Claude response or a silly ‘Hello ‘would have used what was left.

The work had to pause for a few hours. Then continued. (I still resist going for the Max plan here, which would allow larger usage data.)

Pro plan users receive approximately 45 messages in a five-hour rolling window. Large-context tasks eat into that faster than most people expect.

A well-built Skill cannot expand that budget. But the tokens it saves on setup go toward actual tasks. It is an efficient approach for everyone who cannot or doesn’t want to afford a $100/month Max plan.)

Also, the constraint becomes more noticeable the more you try to accomplish in one session.

Thus, a SKILL eligibility filter matters more because of it, not less.

How to apply the rule to prompts you already have

The eligibility rule is worth running against any saved-prompt library you already have.

The full breakdown of what qualifies is in ​Anthropic’s official Skills documentation​. Here is the practical version.

1. Recurring

Do you use this prompt more than once a week, in roughly the same way? A prompt used only for specific one-off projects stays where it is.

2. Structured

Does the task have a fixed shape — same input type, same output format, same constraints each time?

A task that shifts significantly across uses lacks the fixed shape a Skill requires.

3. Consistent output

Does Claude produce results you trust without heavy editing most of the time?

Output that varies in a saved prompt will vary just as much in a Skill. The filter exists precisely for this reason.

Two criteria met: keep it as a saved prompt.

All three met: it belongs in the skills folder.

Practical entry point

Claude Desktop and Claude Code ship with a skill-creator meta-Skill already installed. It runs an interactive Q&A and builds the SKILL.md file automatically.

No manual file writing required. Anthropic also maintains an ​open-source skills repository on GitHub​, featuring production examples spanning document creation, design, and enterprise workflows.

The reliability difference builds quietly. Updating a Skill once updates every future session that uses it. There is no need to hunt through a document for each version that needs changing.

But,

Skills are available to users on the Pro, Max, Team, and Enterprise plans. Free plan users do not have access. Worth knowing before building a library.

Before the next edition

Before the next edition on this topic, I’d love to know where you are with AI, skills, and automation, if that is of any interest to you.

The answers shape what comes next. If most readers are new to Skills, the next edition could be a setup walkthrough.

If most are already building, it becomes a skill-sharing session.

Have you heard of Claude Skills before this edition? Yes / No / Heard of them, never tried

Where do you store your prompts right now? Notes / Notion / Text Blaze / Nowhere organized / Already using Skills

Have Claude session limits ever interrupted a workflow mid-task? Yes, regularly / Yes, occasionally / Not really

Reply directly to this email if you prefer. Either works.

Cheers,

Mark
The AI Learning Guy​
👋⚡😎

Interesting Sources

• ​Grow with Alex — “Claude Skills for Social Media. Build Once, Use Forever.” YouTube, May 11, 2026​
• ​Alex’s free vault — full skill files including video generator, thumbnail generator, and face lock​
• ​The Neuron Daily — “Claude is now the #1 business AI.” May 14, 2026​
• ​Anthropic — Agent Skills official documentation​
• ​Anthropic — Open-source Skills repository, GitHub​

Note: No single website has all the answers. This list serves as a starting point for those who want to explore or satisfy their curiosity about AI. Links: Links with * are affiliate links. See disclosure below.