Skip to content
M.D.N Tech
M.D.N Tech

Claude Code Skills: What They Are and Which Ones to Install

Every skill is written out on the page itself — what it does, who wrote it, and the author's own repository to install from — with a one-command install for the two we publish ourselves.

Browse the skill directory →The skills on GitHub

What a Claude Code skill actually is

A skill is a Markdown file on your machine. That single fact explains most of this page: why there is no registry, no install counter, no account, and why every card in the directory below sends you somewhere else.

A file, and the slash command it registers

A skill is a SKILL.md file living in ~/.claude/skills/<name>/. Claude Code reads the directory, registers a /name slash command, and follows the instructions in the file when you invoke it. There is no plugin runtime, no account, and no network call involved in using one — and no registry to publish a bare SKILL.md to, though packaged plugins do have an official marketplace. The model is reading a file you can open in any editor.

Anthropic documents the format in the Claude Code skills reference. If you are new to the tool itself, start with our complete guide to Claude Code.

Personal scope, project scope

~/.claude/skills/ is personal: those skills are available in every project on your machine. A repository's own .claude/skills/ ships with the codebase, so the skill travels with the work and every contributor gets it on clone. This site's repo carries three that way — handoff, build-kb and test — and its CLAUDE.md instructs every session to open with /handoff start.

Four things that get conflated

SKILL.md
A Markdown file of instructions plus the slash command that invokes them. Local, inert until you type the command. How to install one ↓
CLAUDE.md
Always-on project context, loaded every session whether you ask for it or not. Not a command — a standing brief. How we use it here ↓
MCP server
A separate service Claude connects to over a transport so it can reach systems a local file cannot. Usually needs credentials. Setup commands ↓
Hooks
Shell scripts the harness runs on lifecycle events — session stop, prompt submit — with no model involvement at all. The auto-wrap hooks ↓

Skills add no cost on top of whatever Claude Code plan you already have — they are files the model reads, not a subscription feature someone switches on for you. Claude Code itself is the paid part: it needs a Pro, Max, Team or Enterprise plan, or an API account, and the free Claude.ai tier does not include it.

How to install a Claude Code skill in one command

There is no package manager for skills. Installing one means putting a Markdown file in a directory — which is exactly why every card in the directory below sends you to the author instead of installing for you.

The pattern, for any skill by any author

  1. Make the directory: mkdir -p ~/.claude/skills/<name>.

  2. Save the author's SKILL.md inside it. One file is a complete skill.

  3. Open Claude Code in any project.

  4. Type /<name> to verify it registered.

bash / zsh
mkdir -p ~/.claude/skills/my-skill

That is all any Claude Code skill install is. There is no claude skills install, no package manager and no central registry for bare SKILL.md files — a few catalogue entries ship as packaged plugins through the official marketplace instead — which is why the directory below links out to each author's repository rather than installing anything on their behalf.

The one paste for the M.D.N Tech skills

Read this first: the copy step overwrites same-named files inside your existing skill directories without prompting — and files an older version left behind survive the merge. If you already have skills, back them up before you paste.

bash / zsh — backup
cp -r ~/.claude/skills ~/.claude/skills.bak

Showing the bash / zsh variant.

bash / zsh
git clone https://github.com/bugiiiii11/handoff.git && \
  mkdir -p ~/.claude/skills && \
  cp -r handoff/skills/* ~/.claude/skills/

Verify by opening Claude Code in any project and typing /handoff start.

What it writes

A git clone in your current directory, plus files under ~/.claude/skills/ — nothing else. The clone's skills/ directory is copied in — that is handoff and build-kb, each a single SKILL.md. No daemon, no telemetry, no network call after the git clone. That promise covers these two skills — it is not a claim about the rest of the catalogue, and the optional hooks are a separate story.

Inspect before you run it

Clone the repo, read handoff/skills/handoff/SKILL.md — it is plain Markdown, and it is the whole skill — then copy the folder yourself. You never have to run a script you have not read.

clone only
git clone https://github.com/bugiiiii11/handoff.git

Updating

Pull the clone and re-copy — the command below does both. Re-running the install command only works after you remove the old clone: git clone refuses to write into a directory that already exists.

bash / zsh
cd handoff && git pull && cp -r skills/* ~/.claude/skills/

Uninstalling

Two removals: the skill directories under ~/.claude/skills/, and the clone the install step left in your working directory. After that there is nothing else to unwind.

bash / zsh
rm -rf ~/.claude/skills/handoff ~/.claude/skills/build-kb && \
  rm -rf handoff

Inside the directory: every skill, its author, and where it comes from

A curated directory, not our library. 18 entries from 12 authors: 6 are Anthropic's, 2 are written by M.D.N Tech, and the remaining 10 come from nextlevelbuilder, AgriciDaniel, Corey Haines, multica-ai, obra, Vercel Labs, Firecrawl, Julius Brussee, Trail of Bits and ra100.

What this list is, and what it is not

Every entry links out to the author

We do not host, mirror, re-license or install third-party skills. You install from the author, under the author's terms, from the author's repository.

We carry no licence data

There is no licence field in this catalogue, so we will not tell you a third-party skill's terms — read the repo. MIT applies to the two M.D.N Tech skills published in our repo, and that LICENSE is linked.

Third-party claims are attributed, never adopted

Where a description quotes a figure — a token reduction, a palette count, a number of rules — that is the author reporting on their own work. We did not measure it and do not restate it as ours.

Session management

1 skill

  • Handoff

    Full session lifecycle in one skill: /handoff start, wrap, save, docs. Optional auto-wrap hooks included.

    • Session briefing
    • Context persistence
    • Checkpoint before long context degrades answers
    by M.D.N TechSource for Handoff

Marketing

2 skills

  • Build KB

    Scan your project and generate a chatbot-ready knowledge-base.md in one shot.

    • Chatbot knowledge base
    • Content audit
    • ChatKit onboarding
    by M.D.N TechSource for Build KB

  • Marketing Skills

    Marketing skill collection for AI agents — SEO, CRO, copywriting, paid ads, growth.

    • Conversion optimization
    • Copywriting
    • Growth engineering
    by Corey HainesSource for Marketing Skills

Design

3 skills

SEO

1 skill

  • SEO Audit

    Full website SEO analysis. Crawls pages, detects issues, generates action plan.

    • Website optimization
    • Competitive analysis
    • Ranking improvement
    by AgriciDanielSource for SEO Audit

Infrastructure

2 skills

  • Claude API

    Build Claude API applications. Prompt caching, tool use, streaming integration.

    • API integration
    • AI applications
    • Model optimization
    by AnthropicSource for Claude API

  • Firecrawl

    Web scraping, search, and browser automation for agents at scale.

    • Web scraping
    • Site crawling
    • Real-time research
    by FirecrawlSource for Firecrawl

Safety & validation

1 skill

  • Karpathy Guidelines

    CLAUDE.md guidelines derived from Andrej Karpathy's observations on LLM coding pitfalls.

    • Reduce overcomplication
    • Prevent dead code
    • Clearer goal scoping
    by multica-aiSource for Karpathy Guidelines

Development

3 skills

  • Superpowers

    Composable dev workflow: brainstorm, plan, subagents, and test-driven development in one framework.

    • End-to-end workflow
    • Subagent orchestration
    • Test-driven development
    by obraSource for Superpowers

  • Code Simplifier

    Cleans up recently changed code for readability without altering behavior.

    • Readability cleanup
    • Convention alignment
    • Post-edit polish
    by AnthropicSource for Code Simplifier

  • MCP Builder

    Guided workflow for building high-quality MCP servers.

    • MCP server creation
    • Tool integration
    • API wrapping
    by AnthropicSource for MCP Builder

Testing & QA

1 skill

  • Webapp Testing

    Drive a real browser via Playwright to test local web apps end-to-end.

    • End-to-end testing
    • UI verification
    • Regression checks
    by AnthropicSource for Webapp Testing

Productivity

1 skill

  • Caveman

    Strips narration and filler to cut Claude Code output tokens by ~65%.

    • Token reduction
    • Cost control
    • Faster iterations
    by Julius BrusseeSource for Caveman

Security

1 skill

Documents

1 skill

3D & creative

1 skill

  • Blender Skills

    Expert Claude Code skills for Blender 5.x -- geometry nodes, shaders, rigging, physics, and bpy scripting.

    • Procedural geometry nodes
    • Shader & material recipes
    • Python (bpy) automation
    by ra100Source for Blender Skills

The two skills we wrote and publish

2 of the 18 entries above are ours, and they are the ones you can obtain: Handoff and Build KB. They are MIT, they live in this site's own .claude/skills/, and they are what this page was written with.

Handoff — one skill, four subcommands

Session continuity, with the guarantee that makes each mode trustworthy attached to it.

/handoff start

Reads handoff.md plus git status, the last few commits and the unpushed count in parallel, then briefs the session. Read-only, with exactly one exception: it deletes an emergency snapshot after folding it into the briefing. It deliberately never opens the archive or CLAUDE.md — that is stated token discipline, not an oversight.

/handoff wrap

Updates the docs, rotates overflow into handoff-archive.md, then commits locally. It never pushes, and that is enforced in the skill instructions rather than promised in marketing copy: “Do NOT push.”

/handoff save

The fire exit, for when context is about to run out: a handful of tool calls, no commit, written from what is already in memory. The next /handoff start consumes the snapshot and deletes it.

/handoff docs

A documentation refresh with no commit — wrap's update and rotation steps, and nothing else.

Why it does not grow unbounded

handoff.md is held to a hard line cap, and overflow rotates automatically during wrap: more than two session sections, or a file over the cap, moves the oldest sections into handoff-archive.md. The archive is append-only and never trimmed, so history survives while the file a new session actually reads stays small. That is the whole trick — the briefing is cheap because the state file is bounded by design.

Build KB — a knowledge base out of the repo you already have

It scans the user-facing content already in your repository — README, marketing pages, docs, pricing, support — and writes one knowledge-base.md organised into fixed sections.

It never invents content

Categories with no source content in the repo are skipped and reported as skipped, rather than filled with plausible text. An empty section is information; a fabricated one is a liability.

It asks before overwriting your work

If knowledge-base.md already exists with uncommitted changes, it stops and asks first.

The output pastes into ChatKit — or into any other chatbot. It is a Markdown file, not an export format, and nothing in the skill checks whether you have an account here.

We run these on this codebase

Three M.D.N Tech skills sit in this site's .claude/skills/: handoff, build-kb and test. The project's CLAUDE.md opens with an instruction to start every session with /handoff start. Of those, two are in the public repo; test is a project runner with no source link, so it is described here as proof rather than offered as a download. If you want the people behind them, the team that wrote these skills is on the about page.

Auto-wrap hooks: measuring real context instead of guessing

A long session does not fail loudly. Recall gets patchier and answers get vaguer as the prompt grows, and the moment you notice is well past the moment it started. These two optional hooks decide when to checkpoint by measuring the real prompt size, not estimating it.

What it reads

A Stop hook and a UserPromptSubmit hook read the session transcript and sum input_tokens, cache_creation_input_tokens and cache_read_input_tokens from the last assistant message. That total is the exact prompt size of the most recent API call — the system prompt, CLAUDE.md, every loaded skill, every file read this session. Most context warnings guess from message counts or transcript byte size, which is meaningless once the window is a million tokens wide.

The two thresholds ask for different things. At 20% of the window a soft nudge asks for a checkpoint: run /handoff docs, write the state file, commit nothing, keep working. At 22% the hard nudge asks for the full wrap — stop new work and end the session. Splitting them is the point: a soft nudge that ended the session mid-task only ever got ignored, and a checkpoint costs about two tool calls, so it can always be obeyed.

The assumed window defaults to 1,000,000 tokens, and all three values are environment variables: AUTOWRAP_WINDOW, AUTOWRAP_SOFT_PCT and AUTOWRAP_HARD_PCT. Each threshold fires at most once per session, so a hook can nudge you but cannot trap you in a loop. A hook cannot compact or discard context either — it is an instruction to the model, nothing more, so firing one never costs you anything.

What it buys you

A session that ends deliberately instead of degrading. State reaches the file while the model is still sharp enough to summarise it well, and unfinished work lands in the next-steps table as something resumable rather than a note saying “continue this”. The thresholds are low on purpose: they mark the point where wrapping is still cheap, not the point where the window is full.

What it costs you

Two dependencies and a manual step. The hooks are shell scripts: they need jq and bash, which on Windows means Git Bash. They are opt-in and separate — you copy the .sh files into your project's .claude/hooks/ and register them in your own settings.json. The skill install wires up nothing. “No dependencies” is true of the SKILL.md files and not of these.

bash — copy the scripts
mkdir -p .claude/hooks && cp handoff/hooks/*.sh .claude/hooks/
.claude/settings.json — register them
{
  "hooks": {
    "Stop": [{ "hooks": [{ "type": "command",
      "command": "bash \"$CLAUDE_PROJECT_DIR/.claude/hooks/auto-wrap.sh\"", "timeout": 10 }] }],
    "UserPromptSubmit": [{ "hooks": [{ "type": "command",
      "command": "bash \"$CLAUDE_PROJECT_DIR/.claude/hooks/context-warn.sh\"", "timeout": 10 }] }]
  }
}

Claude Code MCP servers, with the exact setup command

A skill is a local Markdown file that changes how Claude works. An MCP server is a separate service Claude connects to over a transport so it can reach systems a file cannot — which is why skills need no network and MCP servers usually need credentials.

  • GitHub MCP

    Let Claude read issues, review PRs, and search across your repos without leaving the terminal.

    Claude reads the issue and the PR diff itself, so a review starts from the code rather than from your summary of it.

    • Repos & Search
    • Issues & PRs
    • Actions
    Setup command for GitHub MCP
    claude mcp add
    claude mcp add --transport http github https://api.githubcopilot.com/mcp

  • Context7 MCP

    Version-aware library docs and API signatures, injected on demand -- fewer wrong-API errors.

    Targets the most common failure on a fast-moving dependency: an API that changed after the model's training cutoff.

    • Library Docs
    • API Signatures
    • Version-aware
    Setup command for Context7 MCP
    claude mcp add
    claude mcp add --transport http context7 https://mcp.context7.com/mcp

  • Playwright MCP

    A real browser Claude can drive to verify its own UI changes. Microsoft's official server.

    Pairs with the Webapp Testing skill above — the skill knows how to exercise an app, the server gives it a real browser to do it in.

    • Browser Automation
    • DOM Snapshots
    • E2E Testing
    Setup command for Playwright MCP
    claude mcp add
    claude mcp add playwright -- npx @playwright/mcp@latest

  • Sentry MCP

    Pull production errors, stack traces, and releases straight into Claude's context.

    Turns a production stack trace into context, so the fix starts from the trace instead of a description of it.

    • Error Tracking
    • Stack Traces
    • Releases
    Setup command for Sentry MCP
    claude mcp add
    claude mcp add --transport http sentry https://mcp.sentry.dev/mcp

  • Supabase MCP

    Database, auth, and real-time integration with Supabase -- the stack this project runs on.

    The stack this site runs on. The command carries a YOUR_PROJECT_REF placeholder you have to replace with your own.

    • PostgreSQL
    • Authentication
    • Real-time Subscriptions
    Setup command for Supabase MCP
    claude mcp add
    claude mcp add --transport http supabase "https://mcp.supabase.com/mcp?project_ref=YOUR_PROJECT_REF"

  • Blender MCP

    Drive Blender from natural language -- Claude builds, edits, and renders 3D scenes in your viewport.

    Pairs with the Blender skills above: the skills know the bpy API, the server drives the scene that is actually open.

    • Scene & Object Control
    • Materials & Rendering
    • PolyHaven / Sketchfab Assets
    Setup command for Blender MCP
    claude mcp add
    claude mcp add blender -- uvx blender-mcp

None of these are ours

These servers belong to GitHub, Context7, Microsoft, Sentry, Supabase and the Blender community, each published under its own terms and supported by its own maintainers. Nothing connects through M.D.N Tech: the command runs on your machine, the credentials go to the service, and no account here is involved at any point.

Two things worth knowing before you paste. The Supabase command contains a YOUR_PROJECT_REF placeholder that must be replaced with your own project reference. And Firecrawl, listed in the skills directory above, needs its own API key — the “no network calls” property of a local skill stops being true the moment a skill or a server reaches the internet on your behalf.

Who these skills are for — and who they are not for

Four situations these actually solve, and three where you should close the tab. If none of the first four is yours, the honest answer is that you do not need any of this.

This will pay off if

You lose the thread between sessions

Long project, context resets, and every morning starts by re-explaining the same architecture. Handoff briefs the session from a bounded state file and closes it with a commit it will never push, and the auto-wrap hooks decide when to close it for you.

You are shipping solo with no reviewer

Vercel Web Interface Guidelines audits the interface code you just wrote, Webapp Testing drives a real browser against it, Code Simplifier cleans up behind the edit, and Trail of Bits Security Skills is there for the days the stakes justify a security pass.

Your Claude Code bill climbs on long sessions

Caveman strips narration and filler from responses — the author reports roughly a 65% cut in output tokens — and the auto-wrap thresholds keep the prompt side small, which matters because every request re-bills the entire prompt. Two different halves of the same bill.

You need a chatbot knowledge base from a codebase

Build KB reads the user-facing content already in your repo and writes one knowledge-base.md. Paste it into ChatKit, our AI chatbot for websites, or into anything else that accepts Markdown.

Skip this if

You use a different assistant

Cursor, Windsurf, Copilot and the rest are out of scope. Everything here is Claude Code specific: skills install to ~/.claude/skills/ and MCP setup runs through the claude mcp add CLI. There is no adapter and we are not writing one.

You want to install the lot at once

There is no command that installs the catalogue. Ours copies our own skills; every third-party entry installs from its author, on the author's terms.

You pick tools by popularity

We publish no install counts, no stars and no ratings, because we do not collect any. If that is your selection criterion, this list will frustrate you — the objections below explain why we would rather frustrate you than invent a number.

The whole directory is a list, and a list is only useful if it is short and someone stands behind each line.

What this costs, and what M.D.N Tech charges for

The directory is free, so the fair question is what pays for it. Here is what the product that does actually charges.

ToolKit: nothing, and nothing metered

This is a structural fact rather than a promise. There is no ToolKit entry in the billing source of truth at all — no price constant, no credit cost, no trial counter, and no metering call anywhere in the page tree. No account, no email, no card, and no analytics script, install counter or tracking anywhere on it. For a signed-out visitor, the only server call on the live directory decides whether the top bar shows “Login” or an account menu.

So: free today, with no billing code path that could change that quietly. That claim is scoped to ToolKit and to nothing else M.D.N Tech builds.

What we do charge for: ChatKit

Every new chatbot starts with 30 free messages and no card, and an account includes one chatbot. After the trial, replies are metered at one credit per reply, and credits are priced in packs starting at $29, with no subscription. Payment is not live yet — no card is charged today — and the full disclosure sits with the price list on the ChatKit page.

The full price list — every pack and every optional one-time unlock — lives on one page: see what ChatKit costs, including what a chatbot can and cannot do before you pay anything.

Credits are per chatbot, and ChatKit only

Worth stating plainly, because the opposite is the easiest thing to imply: credits are held against an individual chatbot, they are spent only on ChatKit replies, and ToolKit never touches them. There is no shared wallet across products today. One account, not one balance.

If you want the wider picture, the homepage lists all five M.D.N Tech products, shipped and unshipped, with the same separation between what exists and what is still being built.

The objections a sceptical developer should raise

Written as self-interrogation, because a marketing-allergic audience is right to assume the page is arguing for something. Each of these is answered against the code rather than around it.

This is a lead magnet.

It is a page that costs us hosting and earns nothing directly, so the suspicion is reasonable. What it is not is a gate: no account, no email capture, no modal, no interstitial, nothing to dismiss before you read. The ChatKit prices quoted on this page are disclosure, not a checkout — no CTA here leads to a payment, and the skills work identically whether you ever click through or not. If that ever changes — a signup wall, an email-for-download step — the page stops being what it says it is, and the copy above stops being true.

What does the green check on a card mean?

On the live directory in the portal, each card carries a green check. It means editorial selection: a skill we read, evaluated and would use. It is not a verification pipeline — there is no test run, no timestamp, no published criteria, and the flag is set on every entry in the catalogue without exception. A badge that is true of everything tells you nothing, which is why this page does not render it at all. The selection is the signal; the icon was decoration.

How many people use these?

We do not know. There is no telemetry in the skills, no analytics script on this site, no install counter and no GitHub-stats fetch anywhere in the codebase — so there is no number for us to quote, and inventing one would be trivial and undetectable. That cuts both ways: it is also why we cannot show you social proof. Any directory that does publish install counts should be asked where the figure came from and what it counts.

Is it MIT?

The M.D.N Tech skills are, and the LICENSE is in the repo where you can read it. For everything else we hold no licence data at all: the catalogue has no licence field, so we will not tell you a third-party skill's terms — open the author's repository and check. That is the honest answer and it is also the directory's biggest current gap.

What does it do to my machine?

Our install leaves a git clone in your current directory and writes the skill files into ~/.claude/skills/ — nothing else. It overwrites same-named skill directories without asking, which is why the backup command sits above the install command rather than below it. There is no daemon, no telemetry and no network call after the git clone, and the uninstall commands above remove both the skills and the clone. All of that is scoped to our two skills: a third-party skill is the author's code and runs under your review, not ours.

Will you keep adding to it?

No cadence is promised. The catalogue is a hand-edited array in the codebase — no CMS, no feed, no scheduled job — and it changes when we find something we would actually use. We are not going to claim weekly updates to look alive: a directory that grows on a schedule is optimising for looking maintained rather than for staying short.

Why does a chatbot company publish free developer tools?

The straight answer: Build KB turns a repository into a knowledge base, and ChatKit is where some readers will want to paste one. That is the whole commercial logic. The skills work fine if you never do — there is no upsell inside them, no prompt to sign up, and nothing in the code checks whether you have an account. A tool that nags is a tool people uninstall.

Claude Code skills: the questions we actually get

Everything below is answered somewhere in the sections above, in more detail. This is the version for people who arrived with one specific question.

Where do Claude Code skills live on my machine?

Personal skills live in ~/.claude/skills/<name>/SKILL.md and are available in every project you open. Project skills live in the repository's own .claude/skills/ directory and travel with the codebase, so everyone who clones it gets them. Claude Code reads both.

What is the difference between a Claude Code skill and an MCP server?

A skill is a local Markdown file that adds a slash command and tells Claude how to do something. An MCP server is a running service Claude connects to over a transport so it can reach a system it otherwise cannot — your issue tracker, a browser, a database. Skills need no network and no credentials; MCP servers are network services and usually need both.

Do I need a paid Claude plan or an API key?

The skills add nothing on top of whatever you already pay for Claude Code — they are files it reads, not a subscription feature. Claude Code itself does require a Pro, Max, Team or Enterprise plan, or an API (Console) account; the free Claude.ai plan does not include it. Individual entries can have their own requirements — Firecrawl needs a Firecrawl API key, and every MCP server authenticates to its own service — but that is the third party's requirement, not ours.

Can I install one skill instead of the whole directory?

Yes, and it is the only option. There is no command that installs the catalogue. Ours copies the two M.D.N Tech skills out of one repository; each of the other 16 entries installs from its own author's repository, under that author's instructions. Browse the M.D.N Tech skills on GitHub.

What exactly does the install command write, and how do I undo it?

A git clone of the repository in your current directory, plus one directory per skill under ~/.claude/skills/, each containing a single SKILL.md — nothing else. Back up first with cp -r ~/.claude/skills ~/.claude/skills.bak, because same-named directories are replaced without prompting. Undoing it means removing both: rm -rf ~/.claude/skills/handoff ~/.claude/skills/build-kb for the skills and rm -rf handoff for the clone — or Remove-Item -Recurse -Force on Windows.

How do I check a skill is safe before I run it?

Clone the repository and read the SKILL.md. It is plain Markdown with no build step and no hidden code, so reviewing a skill takes about as long as reading a page of documentation. The manual install path is documented above precisely so you never have to run a script you have not read — and that applies to third-party skills too, where the author's code is the thing you are reviewing.

Are these skills licensed for commercial use?

The M.D.N Tech skills are MIT, so yes for those. For every other entry we hold no licence data at all — the catalogue has no licence field — so check the author's repository before you ship with one. We would rather say we do not know than guess on your behalf. Read the MIT LICENSE.

Why do the auto-wrap hooks nudge at 20% of the context window?

For answer quality, not for price — there is no long-context surcharge on current Claude models, which bill one flat rate across the whole window. What does change is recall: the longer the prompt, the vaguer the answers, and by the time you notice it has been happening for a while. The nudge is early so the state file gets written while the model is still summarising well. The two thresholds do different things — 20% asks for a checkpoint that writes the state file and keeps working, 22% asks for the full wrap that ends the session. Both thresholds and the assumed window size are environment variables, so you can move all three.

Do these work on Windows?

The skills do, and the install section has a PowerShell variant of every command. The optional auto-wrap hooks are shell scripts, so they additionally need bash and jq — Git Bash covers both. Nothing else on the page is platform-specific.

Does M.D.N Tech track who installs these skills?

No. There is no telemetry in the skills, no analytics script on this site, no install counter and no account involved in getting them. That is also why we publish no adoption numbers anywhere on this page — we do not have any. Read our privacy policy.

How do I turn my codebase into a chatbot knowledge base?

Install Build KB and run it. It reads the user-facing content already in your repository and writes one knowledge-base.md organised into fixed sections, skipping any section the repo has no source for rather than inventing one. Paste the result into any chatbot you like, including ChatKit, where a new chatbot gets 30 free messages and asks for no card.

Take what is useful, ignore the rest

18 skills listed on this page, each linking to the author who wrote it. Nothing to sign up for, nothing to cancel, and no command that makes you take the whole list.

Browse the skill directory →The skills on GitHub

Built a knowledge base with Build KB? See what ChatKit does with it. Otherwise, there is more engineering writing where this came from.