🤖 Claude’s Most Annoying Habits? These 5 Free Plugins Fix Them Right Away!

Introduction

In this guide, I’ll show you 5 free and open-source Claude Code plugins that many people still don’t know about.

I’ll explain what each plugin does, why it’s useful, who should use it, and show you real example.

I. Claude Code Plugin 1: Destructive Command Guard

This one isn't optional if Claude Code has access to real files. It's the first plugin you should add, full stop.

Key points

• Blocks risky Claude Code commands before they run.

• Works through a separate hook without extra tokens.

• Best for file access, cleanup, client repos, and important projects.

1. What It Is

Destructive Command Guard is a safety hook that sits between Claude Code and your shell. Before Claude Code runs any command, the hook checks it against a blocklist of dangerous patterns. If something matches, it stops the command, tells Claude Code why it was blocked, and waits for you to approve it manually.

It's written in Rust and uses SIMD-accelerated pattern matching, which means the latency is sub-millisecond, you won't notice it's there until it saves you from something catastrophic.

As of July 2026, it has 3.2K GitHub stars and covers 50+ security packs including git, filesystem, PostgreSQL, Docker, Kubernetes, AWS, GCP, and Azure. It works with Claude Code, Codex CLI, Gemini CLI, Cursor, and Copilot.

💡 Install it: 

cargo install --locked --git https://github.com/Dicklesworthstone/destructive_command_guard --tag v0.7.6 destructive_command_guard

2. Who It's For

Here's the thing: Claude Code can misunderstand a request. Especially when you ask it to clean up files or remove something from a project, it might pick a much stronger command than you intended.

Destructive Command Guard catches commands like rm -rf, git reset --hard, DROP TABLE, kubectl delete namespace production, and docker system prune before they run. You can reject it, approve it, or run it yourself when you actually do want to delete something.

One honest caveat though: a Hacker News commenter flagged that Claude Code sometimes works around shell restrictions by writing a temp script and running bash tmp.sh instead. The hook can't always catch that. So this is a safety layer.

II. Claude Code Plugin 2: CanvasUI

Claude Code can build a front end that works well, but the visual style may still feel similar to many other AI-built apps. CanvasUI helps you fix that problem without changing the whole layout.

1. What It Is

CanvasUI is a free, open-source component library with 24 WebGL effects that run over your live HTML. It uses the html-in-canvas browser API to let shaders sample and distort your real DOM in real time, while the content underneath stays fully interactive (text stays selectable, links still work).

Some of the effects included:

• Liquid - fluid simulation that follows the cursor

• Glass - real-time glass distortion over content

• Cloth - physics-based cloth draping

• VHS - scan lines and signal noise

• Particle Reveal - content revealed through particle systems

• Blaze, Shatter, Ripple, Droplets and 15 more

It's MCP-ready, so you can browse and install components directly from Claude Code with a prompt like "add the Liquid component from Canvas UI."

💡 Try it: canvasui.dev, or ask Claude Code directly: "add the Liquid component from Canvas UI"

2. Important Catch Nobody Mentions

I need to be upfront here: CanvasUI requires Chrome with an experimental flag enabled (chrome://flags/#canvas-draw-element). The html-in-canvas API is still experimental. For public-facing sites, you'll need to apply for an origin trial token, and the effects simply won't appear in other browsers.

So this isn't a drop-in for a production site you need to work everywhere. It's best used for demo environments, Chrome-first projects, or landing pages where you control the browser context. Worth knowing before you start building around it.

3. Who It's For

III. Claude Code Plugin 3: Animated UI Components

Your interface can look great in a static screenshot and still feel unfinished the moment someone uses it.

• The loading state is just a spinner.

• The main button has no feedback when clicked.

• The action happens, but nothing tells you it happened.

These small moments matter more than people think.

1. What It Is

Animated UI Components isn't a single repo, it's a category of ready-made interactive state components that Claude Code can use or adapt.

The main components include:

• Thinking Orbs - animated orbs that signal an AI app is processing. Better than a spinner for tasks that take more than a second.

• Border Beam - a light effect that moves around the border of a button or card, drawing attention to a CTA or active state

• Liquid Metal - a flowing metal shader for areas that need strong visual focus

These components exist across several libraries including Magic UI (150+ components), React Bits (90+ components), and individual skill collections. They're all copy-paste style.

You can point Claude Code at a specific library (like Magic UI's shadcn registry) and say "add Border Beam to this button", it'll pull the component and adapt it to your current interface.

2. Why They're Useful

Loading states in most AI apps only tell users one thing: "wait." When a task takes longer than expected, the interface feels frozen because there's no visible change.

Thinking Orbs handles this problem with movement that fits actions such as searching, processing, or writing content. Users still need to wait, but they can see that the app is still working.

Meanwhile, Border Beam works well for a button or section that needs attention, while Liquid Metal gives important actions a stronger visual effect.

Hirbod used a Liquid Metal shader while redesigning the EAS dashboard, which shows how this effect can work inside a real product instead of only appearing on a demo page.

3. Who It's For

IV. Claude Code Plugin 4: No-AI-Slop

Sad to say, but even when Claude Code follows the facts and gets the topic right, the draft can still sound like a robot wrote it. You know the feeling, you read it back and something's off. It's technically correct but... hollow.

No-AI-Slop fixes that with a reusable set of editing rules, so you don't have to paste the same "write more naturally" prompt every single time.

1. What No-AI-Slop Checks

No-AI-Slop is a free, MIT-licensed Claude Code skill built by Peter Yang, a product leader with experience at Roblox, Reddit, Twitch, and Meta. He open-sourced it on July 22, 2026, and it hit nearly 1,000 GitHub stars within a day.

It removes over 20 named AI writing patterns while preserving your original voice. It has two modes:

• Edit mode rewrites the draft, makes the minimum effective change, and shows you a "What changed" summary

• Detect mode finds and quotes every slop pattern it spots without rewriting (for when you want to review before changing anything)

Patterns it targets include:

• Generic openings like "Here's what nobody tells you"

• Forced binary contrasts like "It's not X. It's Y"

• Empty importance-puffery like "It's important to remember" or "It's worth noting"

• Formula closers that restate the whole piece

• Abstract nouns where a concrete number would work better

Critically, it does NOT try to guess whether AI wrote something. It names the specific patterns it finds and quotes the line, so you can verify every edit yourself.

If you already have a humanizer skill, you don't have to replace it. Claude Code can compare both files and add the missing rules to what you already have.

3. Who It's For

V. Claude Code Plugin 5: I-Have-ADHD

1. What I-Have-ADHD Changes

I-Have-ADHD is a free, open-source Claude Code skill from ayghri/i-have-adhd. Developer Jacky shared it on X, and it quickly spread through the developer community.

It runs 10 response rules that reshape how Claude Code structures its answers:

1. Lead with the answer or next action (command, path, or snippet first)

2. Number multi-step work, one bounded action per step

3. End with one next action doable in under two minutes

4. Finish the current issue before raising a new one

5. Restate progress each turn ("step 3 of 5 done")

6. Give time estimates in concrete units, never "a bit" or "soon"

7. After a change, show what now works

8. Errors: state location, cause, and fix

9. Cap lists at 5 items

10. No preamble, no recaps, no closers (unless you ask for an explanation)

💡 Install it: claude plugin marketplace add ayghri/i-have-adhd then type /i-have-adhd to activate. Want it on every session? Run touch ~/.claude/.i-have-adhd-always

Jacky said I-Have-ADHD made Claude's replies "so much better" in practice: important info up front, no unnecessary openers, steps in a clear order.

You can also merge these rules into an existing skill like /quick. Once merged, /quick 3 returns exactly three points, and /quick 5 works when the task genuinely needs more detail. Clean.

2. Who It's For

Conclusion

Start with Destructive Command Guard when Claude Code has access to real files. It's the only one on this list that protects something you genuinely can't recover from.

After that, pick the plugin that matches your actual biggest friction point right now:

• Interface looks too generic → CanvasUI (if you're Chrome-first) or Animated UI Components

• AI writing sounds hollow → No-AI-Slop

• Answers are too long and hard to scan → I-Have-ADHD

They're all free and open source.

If you are interested in other topics and how AI is transforming different aspects of our lives or even in making money using AI with more detailed, step-by-step guidance, you can find our other articles here:

• 5 Best AI Productivity Apps Worth Keeping on Your Phone in 2026

• How to Use Gemini + NotebookLM to Keep Your AI Accurate Every Time (For Free)

• Learn Any New Skill 10X Faster With This Combo (Even Agents Skill, Coding Skill,...*

• Prompting Has Changed. Anthropic’s New Rules with 7 Old Habits You Must Stop!*

• Never Hit a Claude Limit Again with These Methods: From Basics to Advanced Ways*

  *indicates a premium content, if any