Big Claude Code Skill Packs Compared: Jeffallan vs alirezarezvani

If you search for a Claude Code skills pack, two GitHub repos show up again and again: Jeffallan/claude-skills and alirezarezvani/claude-skills. Both promise a huge shelf of ready-made skills, and both are free under the MIT license.

They are built very differently, though. One is a single plugin for developers, and the other is a whole marketplace with more than a hundred plugins.

We cloned both into a throwaway sandbox, loaded them in real Claude Code sessions, measured what they cost your context window, and gave each one the same buggy file to review. This guide shows what we saw and which pack fits which person.

This guide is part of our Claude Code tutorials. Find more add-ons in our Skills Directory.

Tested with Claude Code v2.1.319 on 2026-11-10 (macOS). We used Jeffallan/claude-skills at commit 078a25c (plugin version 0.4.20) and alirezarezvani/claude-skills at commit 8f08254 (marketplace version 2.14.9). The Windows and Linux steps follow Anthropic’s official instructions.

Claude Code Skills Pack Comparison at a Glance

The short version: both packs found the same serious bugs in our demo file, and both cost about the same in context. The real difference is how much of the shelf you want.

What Is a Claude Code Skill Pack?

A skill is a folder with a SKILL.md file that tells Claude how to do one kind of job. A pack is simply many skills bundled together, usually as a plugin.

Anthropic’s plugin docs explain the cost clearly. For each skill Claude can invoke on its own, the name and description sit in Claude’s context on every turn, even in sessions where nothing from the plugin runs. The full text of a skill loads only when it is used.

That is why pack size matters. A bigger pack means more names and descriptions in front of Claude on every single turn.

If you are new to installing add-ons, read our short guide to plugin marketplaces first, or see how Anthropic’s own skills repo loads.

How We Tested Without Touching Our Setup

We never installed anything into our own ~/.claude folder. Both repos went into a sandbox folder, and we loaded them for one session at a time with --plugin-dir and --setting-sources project.

Before loading anything, we read the manifests. Jeffallan ships one plugin.json that points at a skills folder and lists nine project commands. It has no hooks.

The alirezarezvani repo is bigger. Its marketplace file lists 110 plugins, and a handful of them (for example playwright-pro, self-improving-agent and security-guidance) ship their own hooks.json files. We did not load those plugins.

We loaded eight domain plugins whose manifests list no hooks: engineering-skills, engineering-advanced-skills, product-skills, marketing-skills, c-level-skills, ra-qm-skills, finance-skills and business-growth-skills.

We left out pm-skills on purpose. It carries an MCP file that points at an Atlassian server, and we never connect to outside accounts in a test.

Our demo was a tiny Python file, app.py, with an obvious SQL injection, unsalted MD5 password hashing and no login protection. Each session got the same two prompts, and we kept every prompt short.

Jeffallan Skills: One Plugin for Developers

The Jeffallan repo describes itself as 67 specialized skills for full-stack developers. We counted 67 skill folders, and Claude Code listed 67 skills after loading the plugin.

The names read like a list of job titles: python-pro, react-expert, rust-engineer, kubernetes-specialist, secure-code-guardian, code-reviewer, test-master and so on. There are language experts, framework experts, DevOps and security skills.

Each skill in this list shows an estimated cost, from about 80 to 240 tokens for the ones visible in our screenshot. That is the price of having the name and description in context.

The pack also adds nine project workflow commands, such as project:planning:create-epic-plan and project:execution:execute-ticket. Claude Code listed them next to the skills, at about 40 to 50 tokens each. They are written for teams that plan work in epics and tickets, so you may never need them.

Install the Jeffallan Pack

The repo’s own README gives two commands. Run them inside Claude Code.

/plugin marketplace add Jeffallan/claude-skills
/plugin install fullstack-dev-skills@fullstack-dev-skills

Anthropic’s discover plugins page covers the marketplace commands in more detail. The source code lives in the Jeffallan repo.

alirezarezvani Skills: A Marketplace, Not a Plugin

The alirezarezvani repo is a different kind of thing. Its marketplace file lists 110 plugins, and the repo describes skills for engineering, marketing, product, compliance, finance, research and executive advice.

Counts vary depending on where you look. The marketplace text mentions 411 skills, and a plain file count in our clone found 478 SKILL.md files. We cannot tell you which number is right for you, because it depends on which plugins you install.

What we can tell you is what we loaded. Our eight domain plugins gave Claude Code 209 skills.

Every skill name starts with its plugin name, such as c-level-skills:cfo-advisor or marketing-skills:seo-audit. That prefix makes a large list easier to scan.

The descriptions are also shorter. Across the 210 skill files in those plugins, the average description was 221 characters, against 438 for Jeffallan.

Install One Domain at a Time

This pack is built so you pick what you need. The README shows install commands by domain.

/plugin marketplace add alirezarezvani/claude-skills
/plugin install engineering-skills@claude-code-skills

Swap engineering-skills for marketing-skills, product-skills or another plugin name from the repo. You can read the full list in the alirezarezvani repo.

What They Cost in Your Context Window

This is the part most people skip. We opened /context right after loading each pack and read the numbers.

The surprise is that 209 skills cost about the same as 67. In both sessions, many skills in the /context list showed as under 20 tokens each, so the long tail of the list shows very little of its description.

We did not dig into why, and we only tested with a one-million-token window. If you run a smaller window, or the descriptions get trimmed differently, your numbers could differ. Run /context yourself after loading any pack, because that is the only number that matters for your setup.

One thing to remember: both packs used about 1 percent of the window just for the skill list. That is small, but it is paid on every turn. Our guide to reducing token usage shows other places to save.

The Same Review Task, Run Twice

Now the fun part. We asked each pack to use its security skill on app.py, and then its code-reviewer skill. Both packs have a skill with that exact name, which made the comparison fair.

The security prompt for each session was: use the skill, list at most 5 findings in one short line each, and do not edit files.

Security Review Results

The Jeffallan skill found the f-string SQL injection on line 10, the string-built query on line 15, the unsalted MD5 hash, a timing leak from comparing hashes with ==, and missing brute-force protection.

The alirezarezvani senior-security skill found the same two injections, the same weak hashing and the same timing leak. It added a point that the login check can be bypassed through the injectable lookup, and it flagged that duplicate usernames are allowed because the table has no unique constraint.

Honestly, the two lists are close. Both were fast, with Claude finishing in about four to five seconds each.

Code Review Results

The Jeffallan code-reviewer gave three findings: critical SQL injection, critical MD5 hashing, and a major gap, no tests plus a shared global database connection. It closed with a verdict of Request Changes.

The alirezarezvani code-reviewer also led with SQL injection and weak hashing. Its third finding was that the database connection is never closed and that failed writes never roll back.

One practical difference showed up in the prompts. The alirezarezvani skill read two extra reference files, including one for Python.

Claude Code asked us to approve each read, while the Jeffallan skill asked once before it ran. We approved those prompts in our test, and you will likely see similar ones.

This is one demo on one tiny file, run once per pack. It tells you the packs behave sensibly, not which one is smarter overall.

A bigger skill pack is not a smarter one: pick the pack that matches your work, and check /context before you keep it.

Which Pack Should You Install?

Use this ladder. Start at the top and stop at the first yes.

Jeffallan suits you if you write code all day. One command installs the whole thing, and every skill is about software: languages, frameworks, DevOps, security and testing.

alirezarezvani suits you if your work crosses into marketing, product, compliance or finance. You do not install the whole repo. You pick the domain plugins you need, which keeps the shelf small.

If you build websites, you may also want our tested list of web development skills, which includes smaller, focused skills.

Strengths and Weaknesses We Saw

We checked that both repos are active. Jeffallan’s last commit we saw was dated October 30, 2026, and alirezarezvani’s was November 5, 2026.

Try It Yourself

You can repeat our test in a throwaway folder. Clone one pack, load it for a single session and check the cost before you install anything for real.

git clone --depth 1 https://github.com/Jeffallan/claude-skills jeff
mkdir demo && cd demo
claude --plugin-dir ../jeff --setting-sources project

Inside Claude Code, run /skills to see the count and /context to see the cost. Then paste this prompt next to a small file of your own.

Use the secure-code-guardian skill to review app.py for security problems. List at most 5 findings, one short line each. Do not edit or create files.

Before you run any third-party pack, skim its hooks and scripts. Our guide to picking from awesome lists safely has a short checklist.

Common Mistakes

• Installing every plugin in a marketplace at once. Pick the domains you use.
• Trusting the skill count on the README. Run /skills to see what loaded.
• Skipping /context. Skill names and descriptions cost tokens on every turn.
• Approving file reads without looking. Skills can read reference files, so check the path first.
• Assuming a bigger pack is smarter. Our two packs found nearly the same bugs.

Pro Tips

• Call a skill by name, as in “Use the code-reviewer skill”, so Claude does not guess.
• Test a pack with --plugin-dir before you run /plugin install.
• Keep one narrow pack for daily work and add domain plugins only when a task needs them.
• Re-run /context after each new plugin to see its real cost.
• Check the repo’s last commit date before you rely on it.

Cheat Sheet

Frequently Asked Questions

Which Claude Code skills pack is better?

Neither won in our test. Both found the SQL injection and the weak hashing, and both cost about 1 percent of context. Choose by the kind of work you do.

Can I use both packs together?

We did not test them in the same session, so we cannot say how they behave together. Both have skills named code-reviewer, which could be confusing, so call skills by their full plugin name if you try it.

Do these packs cost money?

Both repos use the MIT license and are free to install. Using Claude Code itself still follows your own plan or API billing.

Do skill packs slow down Claude Code?

We did not measure speed. We did measure that skill names and descriptions took about 10k tokens in both sessions, per /context.

Is it safe to install a big skill pack?

Read the manifest and any hooks.json first. The Jeffallan pack and the eight alirezarezvani domain plugins we loaded had no hooks, but a few other alirezarezvani plugins do.

Conclusion

Jeffallan gives developers one tidy plugin with 67 skills, while alirezarezvani gives you a marketplace where you pick only the domains you need. In our test they found nearly the same bugs and cost about the same in context, so let your job decide. Load either one with --plugin-dir first, check /context, and keep only what you actually use. You can also browse more options in our Claude guide and the Skills Directory.