goose-skills

AI Skills for Brand Growth

Put your AI agent on the growth team.

Research customers and competitors, analyze what is working, create the next campaign, and learn from the result. Goose Skills gives Claude Code, Cursor, Codex, and other coding agents 14 marketing playbooks and the video-ad skills behind GooseWorks.

Browse all skills at https://skills.gooseworks.ai

Works with Claude Code · Cursor · Codex

[License: MIT] [Skills]

---

What's in the catalog

Folder	What it holds	Count
skills/marketing/	14 marketing playbooks (one marketer job each, end to end), the 70 steps they load, and 2 foundation skills. Every folder is published	86
skills/video/	Video-ad skills: render atoms, format renderers, mockups, voice, music and the production harness	63
skills/shared/	Skills used across families: the 3 current goose-graphics skills, and 23 older skills that a video skill, a playbook or the GooseWorks app still uses. retired-skills.json lists each one and its user	26
skills/entry/	The entry skills gooseworks install puts in every agent: the gooseworks router, goose-ads, goose-video, make-custom-video, goose-video-local, goose-product-photos. Routing lives in skills/entry/gooseworks/routes.json. The same folder is the Claude Code and Codex plugin (see Entry skills and the plugin)	6

The old GTM library was deleted on 23 Oct 2026. retired-skills.json maps every retired name to the playbook that replaced it (or says why none did), and the old source stays readable at tag legacy-gtm-final.

Contents

• Quick Start
• Brand Growth collection
• Commands
• Repo layout
• Usage Examples
• Building from Source
• Entry skills and the plugin
• Skill Metadata Contract
• Security & Trust
• Agent catalog releases
• License

---

Quick Start

AI Coding Agents (Claude Code, Cursor, Codex, etc)

Paste this into your coding agent and it sets everything up:

Install gooseworks: https://gooseworks.ai/install.md

Your agent reads that page and follows it: it asks you before installing anything, installs the skills and the gooseworks MCP server with the right flag for your agent, opens a browser for Google sign-in, checks the install with npx gooseworks@latest doctor, and (after you restart the agent) offers to set up your company. The page only ever runs npx gooseworks@latest … and touches your agent's skills folder, its MCP config and ~/.gooseworks/credentials.json, nothing else.

Want one skill in particular? Add it to the line: Install gooseworks: https://gooseworks.ai/install.md, including the goose-graphics skill.

After the install, use the skills with /gooseworks <prompt>.

Claude Cowork

Paste the same line into Cowork while working inside a local folder on your machine:

Install gooseworks: https://gooseworks.ai/install.md

Then use the skills like this:

Use /gooseworks skill to generate some ad creatives

Chat apps (claude.ai, ChatGPT)

Chat apps cannot run the install. Add https://mcp.gooseworks.ai/mcp as a custom connector in the app's settings and sign in with Google. The same line above also works: the agent will give you those instructions.

Run it yourself

Prefer the terminal? The same thing, typed by hand (keep --mcp; --all includes it):

npx gooseworks@latest install --claude --mcp   # or --cursor --mcp, --codex --mcp
npx gooseworks@latest install --all            # every detected agent

This gives your coding agent access to the current skills catalog. After installing, just ask your agent to use any skill by name.

If you want a cloud-based AI coworker that already knows all these skills and more, sign up to Gooseworks

---

Brand Growth collection

The Brand Growth collection is the 14 marketing playbooks, grouped by stage. It is not a separate package or command: install GooseWorks once, then ask /gooseworks for the outcome you need.

Stage	What your agent can do	Playbooks
Research	Decide where to start; read the market, competitors and buyers; plan the campaign	where-to-start, research-market, research-competitors, research-audience, plan-campaign
Analyze	See what worked on every channel, run the paid accounts week to week, and show up in search and AI answers	analyze-performance, run-paid-campaigns, improve-search-visibility
Create	Make static ads, content, organic social, email flows, creator briefs and outreach	make-static-ads, make-content, run-organic-social, email-lifecycle, work-with-creators, write-outreach

Video ads and product photos go through the goose-video and goose-product-photos GooseWorks skills, which use the video skills in skills/video/.

Browse the Brand Growth collection

After installation, start with:

/gooseworks onboard me

The agent will collect the useful company context for future growth work and finish by asking what you want to do first. Existing users can skip onboarding and keep using /gooseworks exactly as they do today.

---

Commands

Two command-line tools work with this catalog.

gooseworks is the GooseWorks CLI: it installs the entry skills and the MCP server, and works with your account.

npx gooseworks search "reddit scraping"   # Search the skill catalog
npx gooseworks credits                     # Check your credit balance
npx gooseworks update                      # Update to latest skill version

Installing one skill. This repo is not an npm package. The gooseworks CLI installs any catalog skill, with every skill it requires, straight from this repo (main for production, dev with --env staging):

npx gooseworks@latest install --claude --mcp --with make-static-ads   # or --cursor / --codex

A retired slug is refused with its replacement. Most skills need the GooseWorks MCP server for paid media and live data, which --mcp sets up.

---

Repo layout

skills/
  marketing/<slug>/    marketing playbooks, their steps, foundation skills
  video/<slug>/        video-ad skills
  shared/<slug>/       skills used across families (see retired-skills.json)
  entry/<slug>/        the skills every GooseWorks install starts with; routes.json is the routing source
  entry/.claude-plugin/, entry/.codex-plugin/   the plugin manifests for those same skills
parts/<id>/<x.y.z>/    versioned video building blocks the GooseWorks kit loads (never edited once published)
collections/           curated paths through the catalog (Brand Growth)
docs/marketing/        the marketing build guide and backlog
evals/marketing/       routing and behaviour evals for the playbooks
drafts/marketing/      playbooks built but not released yet
schemas/               JSON schemas: skill.meta.json, and the parts index, layers and withdrawn lists
scripts/, test/        index builder, validators and tests
skills-index.json      the generated catalog every consumer reads
retired-skills.json    retired names, their replacements, and why kept skills stay

A skill's own tests/ folder is for builders: it stays in this repo and is never published to an install.

One flat folder per skill, so a link to another skill is always ../<slug>/SKILL.md. Skill names are unique across the whole catalog and never change: recipes, the GooseWorks app and installed copies refer to skills by name. See CONTRIBUTING.md to add one.

Browse and search every skill at skills.gooseworks.ai.

---

Usage Examples

After installing, just ask your coding agent naturally:

"/gooseworks Make ten fresh static ads for my brand this week"
"/gooseworks What are my competitors running right now, and what works for them?"
"/gooseworks Find 100 heads of marketing at DTC skincare brands"
"/gooseworks Make a video ad for our new serum"
"/gooseworks Use the goose-graphics skill to create a LinkedIn carousel about our launch"

Your agent will search the GooseWorks catalog, download the skill, and run it automatically.

---

Building from Source

git clone https://github.com/gooseworks-ai/goose-skills.git
cd goose-skills
node scripts/build-index.js      # Generate skills-index.json
node scripts/validate-skills.js  # Validate SKILL.md + skill.meta.json contract
node scripts/check-skill-links.js     # Every link in every skill resolves as installed
npm run check:tools              # Every GooseWorks tool a skill names is one the server lists
npm test
npm run ci                       # The checks above in one command (see CONTRIBUTING.md)
npm run check:all                # Every CI job that runs offline (atom checks, part, browser and media tests too)
npm run test:parts               # Part tests (needs ffmpeg; browser tests need playwright-core)

The part tests run on the GooseWorks kit's own ffmpeg build when it is installed (~/.gooseworks/kit/bin/ffmpeg-b<version>), else the one on your PATH; set KIT_FFMPEG to a binary path (or ffmpeg for the PATH one, with KIT_FFPROBE if ffprobe is elsewhere) to choose. Every run prints the ffmpeg it used. The browser tests take PLAYWRIGHT_CORE_PATH (a playwright-core folder) and KIT_CHROMIUM (a Chromium binary). CI sets KIT_FFMPEG=ffmpeg.

---

Entry skills and the plugin

skills/entry/ is published two ways:

• The catalog. gooseworks install and gooseworks update read the entry skills from skills-index.json on main, and the GooseWorks app serves the synced copy.
• The plugin. skills/entry/.claude-plugin/plugin.json and skills/entry/.codex-plugin/plugin.json make the folder a Claude Code and Codex plugin. The GooseWorks plugin marketplace points at this folder on main.

When you change an entry skill, raise the plugin version in both manifests (and the skill's own version). Plugin hosts only update when that number rises. test/entry-skills.test.js checks that both manifests list exactly the entry skills and carry a valid version.

---

Skill Metadata Contract

Each skill folder (skills/<family>/<slug>/) must include:

• SKILL.md — the instructions, with name and description frontmatter (description at most 1,024 characters)
• skill.meta.json — machine-readable metadata

The schema is the contract. schemas/skill-meta.schema.json defines every skill.meta.json field and rejects any other, and node scripts/validate-skills.js checks each meta against it. In brief:

Fields	Who uses them
slug, category (capabilities, composites or playbooks), version, tags	Required for every skill
domain, requires_skills, requires_tools, features, example_prompt, author, collections, collection_stage	Any skill (a shared skill must name its domain)
lint_allow	Video atoms: lint findings that are creative content
tier, playbook, standalone, used_by, pillar, intents, reads, writes, needs	Marketing skills (tier: entry for entry skills)
source_commit, source_hash, source_version	Skills published from goose-studio

Read the schema for each field's meaning and allowed values.

---

Security & Trust

These skills run inside your coding agent, so it's worth knowing exactly what they do:

• Open source & inspectable. Every skill — its SKILL.md instructions and all scripts — lives in this repo under the MIT license. The gooseworks CLI fetches skills at runtime so recipes stay current, but the source you'd run is right here to read, diff, or pin before you run it.
• Where skills land. An install copies each skill into your agent's skills folder (for Claude Code, ~/.claude/skills/<slug>/). When your agent fetches a skill from the catalog in the middle of a task, the gooseworks entry skill tells it to save that skill's scripts under /tmp/gooseworks-scripts/<slug>/, not in your project.
• Where scripts write. Scripts run on your machine and write where they are run or where the agent points them, usually a folder in your project (for example a video's working/ and output/ folders). The shared media helper also keeps two small ledgers under ~/.gooseworks/ (pending-fal-jobs/, so a dropped job is resumed instead of paid for twice, and rejected-fal-requests/, so a refused request is not sent again), and when it has to relay a paid call through your agent it writes the request to working/mcp-requests/. Review any script before letting your agent run it.
• Your agent stays in control. The skills are a tool your agent reaches for when it fits the task (data at scale, sources behind auth, a specific provider) — not a replacement for its built-in web search or fetch on quick lookups. You can read or edit any installed SKILL.md to tune that behavior.
• Credentials stay local. GooseWorks auth is a Bearer token stored at ~/.gooseworks/credentials.json (file mode 0600). Paid media (images, video, voice, music, transcription) goes through GooseWorks' proxies and is charged to your GooseWorks credits, so those skills never need a provider key. All network calls are HTTPS.
• A few older skills can use your own provider keys. They read them from environment variables you set, and only when the GooseWorks route isn't available: reddit-post-finder, linkedin-profile-post-scraper, lead-qualification and job-posting-intent call Apify with APIFY_API_TOKEN when no GOOSEWORKS_API_KEY is set (job-posting-intent also needs your RUBE_TOKEN for its Google Sheets step); launch-meta-ad-campaign calls Meta with META_ACCESS_TOKEN and META_AD_ACCOUNT_ID when the GooseWorks tools aren't connected; brand-research's optional image step reads FAL_KEY from its .env; review-ugc-render falls back to OPENAI_API_KEY, or a local whisper, when the GooseWorks transcription proxy can't be reached. Those calls are billed by that provider, not by GooseWorks.
• The MCP server is part of the install. The install registers the GooseWorks MCP server in your agent's config, because the skills call it for your account. To disconnect it, remove the gooseworks entry from that config (~/.claude.json, ~/.codex/config.toml or ~/.cursor/mcp.json).

Found something that looks off? Open an issue — we'd rather fix it in public.

---

Agent catalog releases

Feature PRs target dev and update staging after CI passes. Promote tested changes to main for production; that switches every customer's entry skills and the plugin too. See DEPLOYMENT.md.

---

License

MIT — see LICENSE for details.

The skill files and scripts in this repository are MIT-licensed. The GooseWorks API they connect to is a separate paid service governed by its own terms.

Built by GooseWorks

Get Started · Report an Issue