Five Claude skills: build your ARSENAL, one task at a time

Commented ARSENAL? Here are the five skills, their complete downloads and a practical way to start. You don't need to know how to code, but you do need Claude Code on a computer, permission to install files, and a willingness to review what the agent asks to do.

A skill is a folder of instructions and sometimes tools that Claude reads for a particular job. It is not a new model, an autonomous employee or a guarantee that an answer is correct. These are community projects, not five products made by Anthropic.

Verification: NOT TESTED BY US. We read the live documentation and downloaded the original files on 21 October 2026. We have not installed these skills in a signed-in Claude session, generated a lesson or diagram, run research, or tested paid integrations. Every installation, prompt and expected deliverable below is documentation-based, not a reproduced result. Manual status: Pendente: teste manual.

Pick one job first

For a first attempt, choose Teach or Marketing Psychology. Keep Chief of Staff for later: it can delegate work and consume more of your Claude usage.

Before installing: your safe practice folder

1. Get Claude Code using the official beginner setup. This guide uses Claude Code on your own computer, not the ordinary Claude chat box. Sign in with an eligible plan or configured API account.
2. Create an empty folder called arsenal-practice in Documents. Open Claude Code in that folder, not your entire home folder, company drive or production website.
3. Download one ZIP below and unzip it. A ZIP does not install itself. Keep the included license and attribution. Ask Claude to explain the skill and its scripts before allowing it to run them.
4. For a project-only install, the destination is arsenal-practice/.claude/skills/NAME/. The file must end up directly at NAME/SKILL.md, not inside an extra nested folder. Create the hidden .claude folder if necessary. You may ask Claude Code to copy the files for you, but approve only that copy first.

Need Claude Code first? Open Terminal on Mac/Linux or PowerShell on Windows. These official installers download and execute Anthropic software; review before approving. Documentation checked on 2026-10-21, NOT TESTED BY US.

Mac, Linux or WSL:

curl -fsSL https://claude.ai/install.sh | bash

Windows, PowerShell only:

irm https://claude.ai/install.ps1 | iex

Reopen your terminal after installation and check the version:

claude --version

Open a terminal in arsenal-practice through your file manager, then start:

claude

On first launch, follow browser authentication and choose your eligible account. If the command is not recognized, reopen the terminal; do not paste unknown repair commands. This guide does not supply a password or sign in for you.

Suggested request, NOT TESTED BY US:

If hidden folders are unfamiliar, use the optional installer commands under Going further instead. Do not paste terminal commands into Claude's ordinary web chat expecting an installation.

1. Chief of Staff: coordinate a goal, not just a chat

By Matt Pocock. This skill explicitly lives under skills/in-progress/: treat it as experimental. It tells Claude to coordinate subagents, use background agents to keep talking with you, and retain dated decisions and evidence in an interlinked Markdown wiki. It suggests recurring schedules only if the host supports them; it does not install a reliable always-on service.

Download here: Chief of Staff: complete skill + MIT license.

Install: unzip; inside chief-of-staff/skills/in-progress/, copy the complete chief-of-staff folder to arsenal-practice/.claude/skills/chief-of-staff/. Keep LICENSE and CAUHI-ATTRIBUTION.txt from the archive with your retained download. Restart Claude Code if the command is not listed.

Try in Claude Code — NOT TESTED BY US:

What to look for: clarifying questions, a plan, a proposed wiki location, and later an index.md linking dated notes. These are intended artifacts, not files we generated. Background agents need supervision; closing the session is not a promise that work continues. A wiki preserves files, not infallible memory.

2. Teach: a course built around your reason to learn

Also by Matt Pocock. Teach starts with your mission, gathers trustworthy resources, and creates short HTML lessons with exercises, printable reference sheets and learning records. It keeps these in the folder where you invoke it, so use the same folder when you return.

Download here: Teach: complete skill, format templates + MIT license.

Install: copy the full teach/skills/productivity/teach folder from the unzipped archive to arsenal-practice/.claude/skills/teach/. Do not copy only SKILL.md: its *-FORMAT.md templates matter. Keep the archive's license and attribution.

Try — NOT TESTED BY US:

What to look for: MISSION.md, RESOURCES.md, lessons/*.html, reference/*.html, learning-records/*.md, plus shared assets/ and notes when needed. Open the lesson in a browser; return to the same workspace for the next lesson. Do not move the HTML away from shared assets without checking its links.

Important correction: the instructions discuss retrieval practice, spacing and interleaving. They do not provide an automatic spaced-repetition scheduler or reminders. Ask for a review exercise and set your own reminder. A generated lesson still needs factual review; it is not an accredited course or professional advice.

3. Anatomy: an interactive explanation, not a stock image

By wheresryan22 / Ryan. Anatomy builds precise isometric figures: SVG for solid parts, WebGL for effects such as water or fire, and a beta mode for parts turning in true 3D. You can request a standalone HTML page, SVG or React components. It is particularly suited to mechanisms, instruments and physical analogies, not arbitrary charts or portraits.

Public project reference, not our hands-on result. Generation: NOT TESTED BY US.

Download here: Anatomy: complete repository, kit, scripts, examples + MIT license.

Install: the archive's anatomy folder is the skill folder. Copy it whole to arsenal-practice/.claude/skills/anatomy/; keep kit/, scripts/, references/ and examples beside SKILL.md. Anatomy requires Node.js 20.10+ for its builds and Chrome/Chromium for browser checks. If you choose the npx skills installer, its separate requirement is Node.js 22.20+ as checked on 21 October 2026.

Try — NOT TESTED BY US:

What to look for: a local HTML file you can open, moving parts and labels tied to the same model. Confirm that the explanation is physically sensible; beautiful geometry is not proof. This is not instant generation: upstream warns of long sessions; it reports 15–20 minutes for a full 3D orbit audit of 30 parts. That is an upstream example as of 21 October 2026, not our timing or a delivery estimate. Start with SVG before trying 3D.

4. Marketing Psychology: choose a principle, then a fair experiment

By Corey Haines. The downloaded skill has 72 third-level principle/model headings, counted on 21 October 2026. This supports “70+,” but is a count of headings, not 72 independent, scientifically validated effects. Some concepts overlap. It connects models such as social proof, anchoring, loss aversion and choice architecture to marketing decisions.

Download here: Marketing Psychology: complete skill + MIT license.

Install: copy the archive's marketing-psychology/skills/marketing-psychology folder to arsenal-practice/.claude/skills/marketing-psychology/. Keep the license and attribution. No additional provider key is documented for this skill itself.

Try — NOT TESTED BY US:

What to look for: a principle, the reason it applies, a concrete ethical change and a measurable test. For example, real reviews may reduce uncertainty; fabricated testimonials are not “psychology.” Do not invent a closing deadline to create urgency. The skill does not publish ads, connect your analytics automatically or prove that revenue will rise.

5. last30days: research recent discussion with visible gaps

By mvanhorn. It searches recent discussion and combines evidence into a sourced brief. The default window is the last 30 days; “recent” does not mean every post is discoverable or indexed immediately. Reddit, Hacker News, Polymarket, GitHub and web are the documented starting sources without optional platform setup; actual web access also depends on the agent's search tools or available backend.

Public documentation reference. Searches, account setup and integrations: NOT TESTED BY US.

Download here: last30days: runtime complete snapshot excluding promotional media, Python engine, references + MIT license.

Install: the archive is a runtime complete snapshot excluding promotional media, not an immediately installable skill folder. Inside last30days/skills/, copy the complete last30days subfolder into arsenal-practice/.claude/skills/last30days/. Check that SKILL.md, scripts/ and references/ are siblings. The engine requires Python 3.12+; optional setup uses additional tools. Keep the repository's root license and attribution with your download.

Try — NOT TESTED BY US:

On first run, choose Skip for now if you do not want optional setup. X needs an authenticated route or provider configuration. YouTube uses yt-dlp for public search/transcripts and needs that tool installed. TikTok and Instagram need a ScrapeCreators API key. Do not approve browser-session access just to finish a tutorial.

Quota warning, checked 21 October 2026: upstream advertises 10,000 free calls, but the provider's public site says 100 starting credits, with up to 7,000 more through conditional rewards. We did not authenticate to verify a partner-specific allowance. Do not budget on 10,000 free calls; check your account dashboard. With a key, Reddit/YouTube backfills and transcript rescue can also spend credits, not just TikTok/Instagram.

Going further: optional commands, configuration and recovery

Everything in this section is NOT TESTED BY US. It is separate from the ZIP route above. Choose one installation method, not both, to avoid duplicate skills. Installer commands retrieve upstream's latest files, which may differ from our pinned downloads.

Install with the skills CLI

Install a current Node.js LTS that meets 22.20+ from nodejs.org. npx comes with npm; a terminal is your computer's command window. Run these in your practice folder's terminal, one at a time, after reviewing the corresponding download. Each command targets Claude Code and stays project-local because -g is omitted.

npx skills add https://github.com/mattpocock/skills/tree/main/skills/in-progress/chief-of-staff -a claude-code

npx skills add https://github.com/mattpocock/skills/tree/main/skills/productivity/teach -a claude-code

npx skills add wheresryan22/anatomy -a claude-code

npx skills add coreyhaines31/marketingskills --skill marketing-psychology -a claude-code

npx skills add mvanhorn/last30days-skill --skill last30days -a claude-code

The direct Chief of Staff path matters: it is nested in an experimental directory. Do not install every skill from a repository by accident. Review installer prompts; do not bypass them with automatic confirmation.

Add -g only if you want a user-wide install across projects, normally visible to Claude Code under ~/.claude/skills/. On Windows, ~ means your user home folder. A project install is easier to isolate. To inspect installed skills or check updates:

npx skills list

npx skills check

Upstream also documents npx skills update and npx skills remove; review the selected scope and names before approving. A manually copied ZIP does not automatically track updates. Retain a backup before replacing files. To remove a manual install, delete only its named skill folder, not your lessons, wiki or research outputs.

Chief of Staff, Teach and Marketing: configure the workspace, not secret switches

Chief of Staff: agree a wiki directory before the first write, require dated decisions with source links, and set approval boundaries for delegation and schedules. The skill does not document a universal background-service configuration. Ask what scheduling capability your Claude Code environment actually provides before relying on recurring work.

Teach: keep one learning folder per topic. Your purpose belongs in MISSION.md, sources in RESOURCES.md, preferences in NOTES.md, and progress in learning-records/. Ask for a mission change explicitly rather than silently changing the curriculum. Resume with /teach Continue from my existing mission and learning records; give me a short recall exercise before new material. This is a suggested request, not a special scheduler command.

Marketing Psychology: give it real product, audience and pricing context. The skill checks .agents/product-marketing.md, then documented Claude/legacy alternatives such as .claude/product-marketing.md. Ask the agent to draft this context for approval using only information you provide. No private customer list is needed. Invoke /marketing-psychology with the page or offer to review; do not invent extra configuration flags or assume installing the skill connects advertising accounts.

last30days: marketplace alternative and settings

Its recommended Claude Code route is the plugin marketplace. These go inside Claude Code, not your system terminal. Do not use this alongside a second manual/CLI installation:

/plugin marketplace add mvanhorn/last30days-skill

/plugin install last30days

For a shorter research window, use /last30days your topic --days=7. --quick reduces depth; --deep increases it and can increase time, usage and API spend. Ask /last30days check source health to request its doctor procedure; Claude Code's separate /doctor is not the same check. Ask to export an existing brief as HTML rather than researching again. Keep it local; publishing is a distinct opt-in and may make it public.

Global settings live in ~/.config/last30days/.env; trusted project settings can live in .claude/last30days.env. Add real keys only through a secure local editor or the provider's authorized setup, never a public chat, screenshot or committed file. Examples below contain no real credentials:

BROWSER_CONSENT=false
FROM_BROWSER=off
AGENTCOOKIE=off
LAST30DAYS_MEMORY_DIR=~/Documents/Last30Days
LAST30DAYS_REDDIT_SC_MIN_ITEMS=0
LAST30DAYS_YT_SC_MIN_ITEMS=0

The first three disable the documented browser-cookie routes; the last two limit keyed Reddit/YouTube search backfill to empty results, not all paid calls. They do not disable transcript rescue or guarantee zero cost. To avoid provider spend, skip keyed integrations entirely. SCRAPECREATORS_API_KEY enables its provider; XAI_API_KEY is one optional X route. Neither is included in a Claude subscription. Ask the agent to explain the active sources without printing secret values.

Saved slash-command research normally goes to ~/Documents/Last30Days; an explicitly empty LAST30DAYS_MEMORY_DIR disables those brief saves. Teach instead writes into its learning workspace; Chief of Staff should agree a wiki path with you. Back up these folders if you want to keep the work.

Anatomy: output and browser options

Ask for “standalone HTML, light theme” first. “SVG only” avoids 3D complexity; “React component” is for an existing developer project, not necessary for beginners. CHROME_PATH can point to a nonstandard Chrome/Chromium installation for checks. Do not disable all sandbox protections to make a screenshot work: ask for the precise failing check and approve only a scoped action you understand.

FAQ: cost, mobile and safety

Is it free?

The five downloads are MIT-licensed, with original notices included. Claude Code usage is separate. On 21 October 2026, Anthropic's pricing page listed Pro at US$20/month when billed monthly, or US$200 paid upfront annually (displayed as about US$17/month), before applicable taxes. Pro includes Claude Code with usage limits; the ordinary Free plan does not include Claude Code. Max started at US$100/month. Console/API billing and third-party API credits are separate routes/costs. Check checkout, exchange rates and your usage settings; these are not unlimited plans.

Can I do this on a phone or in ChatGPT?

You can read this guide, save downloads and view suitable exported pages on a phone. This installation path requires a computer running Claude Code; pasting /teach into ordinary Claude or ChatGPT chat does not install anything. Remote Claude Code interfaces exist, but we have not verified these five in mobile/remote environments. Do not assume file access, browser checks or background execution work the same way.

Do I need to understand programming?

Not to describe a goal or use an exported lesson. You still need to install the host, identify a folder and review permissions. If a command is unfamiliar, ask for an explanation before executing it. Do not give an agent unrestricted access because it promises to handle the technical work.

What if the command does not appear?

Check NAME/SKILL.md and all supporting files, confirm you opened the correct workspace, and restart Claude Code. For the marketplace route, check that the plugin is enabled. Chief of Staff and Teach have manual invocation configured; type their slash commands explicitly. Do not treat a generic answer as proof the skill loaded. Ask which exact skill file was loaded; missing runtime files are an error, not permission to invent results.

Is it safe for client or company data?

No blanket guarantee. Read the instructions and scripts, start in a disposable folder and use fictional data. Never share API keys, browser cookies, tax IDs or customer lists in prompts. Research sends queries to external services; prompts and files may be processed by your agent provider under your account terms. Retrieved posts can contain malicious instructions: treat them as evidence, not commands. Require separate approval before network changes, account access, spending, publication or scheduling. Keep an independent backup.

Can I trust the outputs?

Check citations, dates, arithmetic and mechanisms yourself. Research popularity is not truth; prediction-market odds are not financial advice; a polished diagram can be wrong. These five skills do not guarantee productivity, learning outcomes, conversions or comprehensive social coverage. Manual end-to-end testing remains pending for all five.

Sources and attribution

Original work: Matt Pocock (Chief of Staff and Teach), Ryan / wheresryan22 (Anatomy), Corey Haines (Marketing Psychology), and mvanhorn (last30days). Download archives preserve original MIT licenses and add CAUHI-ATTRIBUTION.txt; Anatomy includes a complete repository snapshot; last30days includes a runtime complete snapshot excluding promotional media (media/pr-assets/); the others include the complete named skill directories and root notices. These are not certifications by their authors or Anthropic.

Read the source ledger, pinned commits, download hashes and pending tests on this site. Evidence checked 21 October 2026. Install/usage behavior: NOT TESTED BY US — Pendente: teste manual.