This is the full guide from my reel on five add-ons for Claude Code. You get the link to each one, the exact install command from its own page, what problem it fixes, and where it falls short. Everything here was checked against each project’s own page on 11 November 2026.

Claude Code is Anthropic’s coding agent: you type what you want in plain words and it writes and runs the code. A plugin or skill is an add-on that teaches it a new trick.

The short version

• Usage: FreeLLMAPI puts the free tiers of 34 AI providers behind one address and moves to the next model when a limit runs out.
• Video: Lemo-Opuscar gives Claude Code 66 ready film styles for animated explainers made fully in code.
• Output: Caveman makes Claude answer in fewer words. The real saving is smaller than the headline number.
• Data: last30days researches a topic across Reddit, X, YouTube, Hacker News and the web.
• Security: NVIDIA SkillSpector scans a skill for malware and hidden instructions before you install it.
• In the reel I called all five plugins. All five are free and open source, but two of them (the first and the last) are separate tools that work with Claude Code, not plugins in the strict sense.

Before you start

• Claude Code needs a paid plan. Anthropic’s pricing page says Claude Code is included in all paid plans and not in the Free plan (claude.com/pricing). Prices are shown in US dollars; check the page for what you pay in India.
• Install Claude Code first. Follow Anthropic’s quickstart.
• Two kinds of commands. Lines that start with a slash, such as /plugin, are typed inside Claude Code. All other lines are typed in your terminal (the Terminal app on Mac, PowerShell on Windows).
• /plugin install opens a panel first. Anthropic’s docs say that inside Claude Code this command does not install straight away. It opens a panel with the plugin’s details. Read what the plugin adds, then choose “Install for you”. If the new commands do not show up, type /reload-plugins.
• Read before you install. Anthropic’s plugin security page says a plugin you install can run code on your computer with your own permissions, and that Anthropic does not control or check what third-party plugins contain. Install one at a time, and scan anything unknown with tool 5 first.

#	Tool	Fixes	Install type
1	FreeLLMAPI	Running out of usage	Separate app + one setup command
2	Lemo-Opuscar	Animated explainer videos	Claude Code plugin
3	Caveman	Long, wordy answers	Claude Code plugin
4	last30days	Fresh data from social platforms	Claude Code plugin
5	SkillSpector	Unsafe skills	Command-line scanner

---

1. FreeLLMAPI: free model usage in one place

Link: https://github.com/tashfeenahmed/freellmapi

Many AI companies give a small free allowance every day. FreeLLMAPI is a small program on your own computer that collects those allowances behind one address. Its README lists 34 providers, 635 free model endpoints and roughly 7.4 billion tokens a month of listed free capacity. (A token is a small chunk of text; AI usage is counted in tokens.) When one model hits its limit, the router retries the next model in your list by itself.

Important: in the reel I called this free AI tokens. To be exact: this does not give you free Claude. It lets Claude Code run on other companies’ free models, such as Gemini, DeepSeek or Kimi, when you do not want to spend your Claude limit. You sign up with those companies yourself and bring your own free keys.

Setup

1. Install the router. On Mac or Windows, the simplest way is the desktop app from the Releases page. If you would rather use Docker (a tool that runs apps in a sealed box), use the Docker installer command below instead.
2. Open the dashboard and go to the Keys page. Desktop app: click the tray icon, then Open Dashboard. There is no login. Docker install: open http://localhost:3001 in your browser. Docker installs have an email and password login for the dashboard.
3. Create a free account with each provider you want and paste its API key on the Keys page. You bring the keys; the tool does not come with any.
4. Copy your unified API key from the top of the Keys page.
5. Run the dry run command below. It only shows what would change in your Claude Code settings. Desktop app users: copy the address shown in your browser when the dashboard opens and use it in place of http://localhost:3001.
6. If it looks fine, run the real run command. It makes a dated backup of your settings first.

Docker installer (step 1, only if you are not using the desktop app). Docker must be installed and running. It is a bash command, so run it in a Mac or Linux terminal:

curl -fsSL https://freellmapi.co/install.sh | bash

Dry run (step 5):

npx freellmapi setup-claude --url http://localhost:3001 --api-key <unified-key> --dry-run

Real run (step 6):

npx freellmapi setup-claude --url http://localhost:3001 --api-key <unified-key>

In both commands, replace <unified-key> (including the angle brackets) with the key you copied in step 4. The npx commands need Node.js 20.18 or newer. To keep your normal Claude setup untouched, add –profile and a name to the real run command; the docs say this creates a separate named profile.

Limits and risks

• The README says it is for “personal experimentation and learning, not production”.
• Its own limits list: free tiers can change or vanish without notice, speed varies a lot, there is no uptime promise, and answer quality drops later in the day as the best models hit their daily caps (they reset at midnight UTC, 5:30 am in India).
• Free installs get new models 30 days late. The README says the free build currently sits about 303 models behind the live list. The live list is a paid extra: $19 a year or $49 once.
• Each provider’s terms still apply to you. The project’s own review marks some, such as Google Gemini, as “caution”.
• Run it only for yourself. The docs say not to expose it to the internet.

---

2. Lemo-Opuscar: animated videos in 66 styles

Link: https://github.com/lemomo-ai/lemo-opuscar · Style gallery: https://lemomo-ai.github.io/lemo-opuscar/

This plugin turns Claude Code into a small video studio. You name a style and a topic, and Claude writes code that draws every frame, adds music and a text-to-speech voice, and exports a finished video. The README says it uses “no video generation, no stock footage”. It lists 66 styles as of November 2026, including Chalkboard, Transit Map, Caption News, watercolor and Swiss Motion Graphics.

Install (in your terminal, not inside Claude Code)

Run these two commands one after the other:

claude plugin marketplace add lemomo-ai/lemo-opuscar

claude plugin install lemo-opuscar@lemolab

Then start Claude Code again in any folder (or type /reload-plugins if it is already open) and ask in plain words. Example prompt:

Make a 45-second vertical (1080x1920) film in the Chalkboard style that explains how UPI works. Warm female narrator. Show me a storyboard first.

Not sure which style to pick? Type this and it will ask you a few questions:

/lemo-opuscar:lemo-opuscar

Limits and risks

• The README says one film takes the agent “about 30–60 minutes and a fair amount of tokens”. It does not say how much of a plan’s limit that is, so make one short film first and watch your usage.
• The styles are tuned for Claude Opus 5.5. The README warns that other models may not reproduce them well.
• Your computer needs Node 20+, ffmpeg 7+ and Python 3.11+. The agent installs the rest.
• Default output is 1920×1080 at 24 frames per second. For Reels or Shorts, ask for vertical.
• On first use it downloads its guides and style files to a lemo-opuscar folder in your home folder.
• Films you make are yours. Six voice tracks in the demo films are not free to reuse, so do not copy audio out of the demos.

---

3. Caveman: shorter answers

Link: https://github.com/JuliusBrussee/caveman

Caveman makes Claude drop the filler: no greeting, no recap, answer first. Code, commands and error messages stay exactly as they are. Shorter answers mean fewer output tokens.

Install (inside Claude Code)

/plugin marketplace add JuliusBrussee/caveman

/plugin install caveman@caveman

The second command opens the plugin panel. Check what it adds, then choose “Install for you”.

It starts by itself in new sessions. If it does not, type /caveman. For even shorter replies, type /ultracave. To go back to normal, type:

stop caveman

The README’s main install is a bigger command-line tool (npx @caveman-ai/cli@latest, needs Node.js 22.13+) that also compresses what Claude reads and can pick models for you. That tool sends usage statistics by default; the README says you can turn this off with the command below. If you only want shorter answers, the plugin above is enough.

caveman telemetry off

How much does it really save?

In the reel I said one benchmark showed more than 50% less output. Here are the numbers behind that, and the ones that disagree:

• The project’s one-line description on GitHub says it “cuts 65% of tokens”. The README itself no longer leads with that number, and the project’s own Honest Numbers page says earlier versions used a fixed 65% figure without a reviewed test behind it.
• The README’s own test on ten developer questions: 6,983 output tokens with no instruction, 4,334 with a plain “Answer concisely.”, 4,119 with /caveman (about 41% less than no instruction) and 2,693 with /ultracave (about 61% less). So more than 50% happens only in the strongest mode, compared with giving no instruction at all.
• Against that plain “Answer concisely.” instruction, the README says /caveman cuts 3% more and /ultracave 35% more at the median.
• JetBrains tested Caveman itself on 86 real coding tasks in Claude Code and measured 8.5% fewer output tokens, with no measurable drop in quality. JetBrains forced the skill on for every reply, so it calls 8.5% the ceiling, not the usual result. Reason: in real coding, most output is code and tool calls, which Caveman leaves alone.
• An outside test listed on the Honest Numbers page (LemonCrow, 300 runs per side) measured 44.1% less billed output, but the total cost changed by +0.06%.
• Elastic tested its own remake of the idea on eight scenarios and measured 63.6% fewer response tokens.

So expect a big cut on chatty question-and-answer use and a small one on real coding work. The README says the rules themselves cost about 1,160 input tokens, and that on tools billed per request (it names GitHub Copilot) shorter answers save nothing.

---

4. last30days: research from Reddit, X and more

Link: https://github.com/mvanhorn/last30days-skill

The README’s pitch is that no single AI tool can search Reddit comments, X posts and the other platforms in one go, and that Claude has none of them built in. This skill searches many platforms at once for what people said in the last 30 days, ranks results by real engagement (upvotes, likes, views) and writes one summary with sources.

Install (inside Claude Code)

/plugin marketplace add mvanhorn/last30days-skill

/plugin install last30days

The second command opens the plugin panel. Check what it adds, then choose “Install for you”. Then type /last30days followed by your topic:

/last30days best AI tools for small businesses in India

What is free and what needs setup

• Works straight away with no keys: Reddit (with comments), Hacker News, Polymarket and GitHub.
• X: you log in to x.com in your browser so the tool can use that session, or you add an API key. The first-run setup wizard walks you through it.
• YouTube: needs a free tool called yt-dlp (command below).
• TikTok and Instagram: need a ScrapeCreators key. Threads, Pinterest and LinkedIn need the same key plus switching them on in the settings. The README says the setup wizard sign-up gives 10,000 free calls; ScrapeCreators’ own site advertises 100 free credits for a new account, with up to 1,200 more to claim, so check what you actually get. After that it is pay as you go.
• Web search: a Brave Search key. The README says 2,000 free queries a month, but Brave’s own page now shows $5 of free credit every month on its search plan and asks for a credit card, so check it before you sign up.

To add YouTube on a Mac that has Homebrew (a free installer for Mac tools), the README gives this command. On Windows, follow the install steps on yt-dlp’s GitHub page.

brew install yt-dlp

Limits and risks

• For X it reads your logged-in browser session. Only do this on your own computer, and only if you are comfortable with that.
• What you get depends on your keys, the tools installed and the topic. The configuration page lists what each source needs.
• It is built with Python 3.12+, so you need that on your computer.
• It tells you what people are saying, not what is true. Check important facts yourself.

---

5. NVIDIA SkillSpector: scan a skill before you install it

Link: https://github.com/NVIDIA/SkillSpector · NVIDIA’s guide: https://docs.nvidia.com/skills/scanning-agent-skills

Anyone can publish a skill, and a skill runs with your permissions. SkillSpector’s README cites research on 31,132 skills in which 26.1% had vulnerabilities and 5.2% showed likely malicious intent. The scanner reads a skill’s files without running them and checks 71 patterns in 17 categories per its README (NVIDIA’s guide page still says 68). These include known malware signatures, crypto miners, data theft and prompt injection (hidden text that tries to give the AI orders).

Install (in your terminal)

It is a command-line tool, installed with uv, a free Python installer. First install uv. These are the commands from uv’s install page. Mac or Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows (PowerShell):

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Close the terminal, open a new one, then install SkillSpector:

uv tool install git+https://github.com/NVIDIA/skillspector.git

Scan a skill

To scan a skill straight from GitHub, use the command below. Replace the link with the GitHub link of the skill you want to check:

skillspector scan https://github.com/user/my-skill --no-llm

For a folder you already downloaded (replace ./my-skill/ with your folder; a .zip file works the same way):

skillspector scan ./my-skill/ --no-llm

The README says remote downloads and zip files are capped at 100 MiB. You get a risk score from 0 to 100:

Score	Severity	Advice
0–20	Low	Safe
21–50	Medium	Caution
51–80	High	Do not install
81–100	Critical	Do not install

Limits and risks

• The –no-llm flag is the fully free mode. File contents stay on your computer. One check still goes online: it sends the names and versions of the packages the skill depends on (not your files) to OSV.dev, a public database of known vulnerabilities.
• Without –no-llm, the tool also sends file contents to an AI provider for a deeper check. That needs an API key, or a local option such as your Claude Code login or Ollama.
• The README calls it “defense-in-depth, not a sandbox”. It never runs the skill, so it cannot see what the skill does at run time.
• Its own limits list says it may miss non-English content, text inside images, and encrypted or compiled code.
• A low score is not a guarantee. Still read what the skill asks for, and prefer skills from authors you can identify.

Do not want to use a terminal? Bitdefender AI Skills Checker is a free web page where you paste a skill’s GitHub link or upload a zip. Cisco’s Skill Scanner is another open-source option; its README also says that “no findings” does not mean no risk.

---

Common problems

• “command not found: claude” Claude Code is not installed, or the terminal was not reopened after installing. Go back to the quickstart.
• “command not found: npx” Install Node.js from nodejs.org, then open a new terminal window.
• “command not found: uv” or “skillspector” Close the terminal and open a new one after installing uv. If skillspector is still missing, run the uv tool install command again and read the last lines it prints.
• A /plugin command does nothing. Slash commands only work inside Claude Code, not in the plain terminal.
• I ran /plugin install and a panel opened. That is normal. Choose “Install for you” in the panel to finish.
• The plugin is installed but its commands are missing. Type /reload-plugins, or close Claude Code and start it again.
• How do I remove a plugin? Type /plugin inside Claude Code and open the Installed tab. Each plugin has Disable and Uninstall options.
• Claude Code feels less smart after FreeLLMAPI. It is now using free models, not Claude. Restore the backup that setup-claude made, or use a separate profile.
• Should I install all five? No. Pick the one that fixes your biggest problem and try it for a week.

Sources

• FreeLLMAPI README, client setup guide and limitations page: https://github.com/tashfeenahmed/freellmapi
• Lemo-Opuscar README: https://github.com/lemomo-ai/lemo-opuscar
• Caveman README: https://github.com/JuliusBrussee/caveman
• Caveman Honest Numbers page: https://github.com/JuliusBrussee/caveman/blob/main/docs/HONEST-NUMBERS.md
• JetBrains test of Caveman, July 2026: https://blog.jetbrains.com/ai/2026/07/speak-to-ai-agents-like-cavemen-tosave-tokens/
• Elasticsearch Labs test: https://www.elastic.co/search-labs/blog/elastic-caveman-ai-token-reduction
• last30days README and configuration page: https://github.com/mvanhorn/last30days-skill
• Brave Search API page: https://brave.com/search/api/
• ScrapeCreators: https://scrapecreators.com/
• NVIDIA SkillSpector README: https://github.com/NVIDIA/SkillSpector
• uv install page: https://docs.astral.sh/uv/getting-started/installation/
• Claude Code plugin docs: https://code.claude.com/docs/en/discover-plugins
• Claude Code plugin security page: https://code.claude.com/docs/en/plugins/security
• Claude plans and pricing: https://claude.com/pricing

All numbers above are the projects’ own claims unless a tester is named. These tools change often, so check each README for the latest commands. This page is general information; install third-party tools at your own risk.