AI Guides › Workbench

The Gather-Then-Make Bench: Seven Claude Skills for a One-Person Creative Studio

By Nigel Guy · 7 min read

The usual way into "an AI creative studio" is to pay someone thousands to wire together a stack you can't maintain, or to install every skill you see in a thread and hope they cooperate. Either way you end up with a video tool fed by guesswork, because nothing upstream did the research. The studio fails at the handover between jobs, not inside any one of them.

The rule: install the skills in the order the work happens. Gather first, then make, and don't add a making skill until the gathering one before it produces something you'd actually use.

What this is, and what it isn't

A skill is a folder with a SKILL.md file of instructions, and sometimes scripts, that Claude loads when a task matches. In Claude Code, personal skills live in ~/.claude/skills/<skill-name>/SKILL.md and you can call one directly with /skill-name. In the Claude apps, skills need Settings > Capabilities > Code execution and file creation switched on, then you add them under Customize > Skills as a ZIP (on Team and Enterprise an admin enables them first under Organization settings > Plugins & skills).

It is not a studio in a box. Skills are instructions, not services. Several below need their own paid API key, one needs a licence once your company grows, and none of them judge taste for you.

The bench

Prices change often. Several vendors list only US dollars, so treat any figure here as approximate and check the £ amount, including VAT, at checkout.

Skill Job Cost at time of writing Best for The catch
competitive-ads-extractor (Composio collection) Pulls competitors' ads from ad libraries and sorts their messaging Free skill; needs a browser tool in Claude Seeing what others are saying before you write It's a workflow, not a scraper. Without browser access it has nothing to read
deep-research (sanjay3290/ai-skills) Runs Google's Gemini Deep Research agent and returns a cited report Free skill; the README estimates $2–5 per task in Gemini API usage Market and audience background Needs GEMINI_API_KEY and Python 3.8+; takes 2–10 minutes
content-research-writer (Composio collection) Outlines, researches, cites and gives feedback section by section Free Turning research into a script or article Works best on a project folder of drafts, not a one-line request
A voice skill you write yourself (with Anthropic's skill-creator) Holds your tone, banned phrases and examples Free Keeping every output sounding like you Only as good as the samples you feed it
ElevenLabs skills (official) Text to speech, sound effects, music, dubbing and more Free tier of 10,000 credits a month, but no commercial use; paid plans start at about $6 a month Voiceover for the finished script Needs ELEVENLABS_API_KEY; free-tier audio can't go in client or paid work
Remotion skills (official) Teaches Claude to build and render videos in React with Remotion Free for individuals, non-profits and for-profit companies of up to 3 employees Turning script and audio into video Larger companies need a paid Company Licence; you need Node.js and a terminal
skill-creator (Anthropic) Interviews you and drafts a new skill Free; included in Anthropic's example skills Building the voice skill above The draft needs a human edit before it's trusted

Six of these do the studio's jobs. The seventh, skill-creator, builds the one piece nobody can download for you.

The three that gather

1. competitive-ads-extractor. Copy the competitive-ads-extractor folder from the ComposioHQ awesome-claude-skills repository into ~/.claude/skills/. It works from the Facebook (Meta) Ad Library and LinkedIn, and asks for screenshots, a CSV of copy and calls to action, and a markdown summary of themes. Give it named competitors, not a category.

You are helping me research competitor advertising before I write anything. Use the competitive-ads-extractor skill.
Competitors: [COMPETITOR_NAMES_OR_PAGE_URLS]
My product and audience: [ONE_LINE_DESCRIPTION]
Steps: 1) collect each competitor's currently visible ads; 2) record copy, offer, call to action and format in a table; 3) group the ads by the problem they lead with; 4) list the gaps nobody is covering.
Output: the table, then the groups, then no more than five gaps.
Rules: only record ads you actually saw, with the date you saw them. Don't guess at spend or results, because ad libraries don't show performance. If you can't reach a library, stop and tell me rather than working from memory.
Before answering, check that every row has a source and a date.

Fill in the competitor names or ad-library page links, and one line on what you sell.

2. deep-research. This one lives in sanjay3290/ai-skills, not the Composio repository. Composio only links to it. Set GEMINI_API_KEY from Google AI Studio, then run it through Claude or directly with python3 scripts/research.py --query "..." --stream. Add --status with the ID it returns to check a long run. If you'd rather not run another API bill, Claude's own Research mode does a lighter version of this job inside the chat. Skip the skill until you need the depth.

3. content-research-writer. Copy the folder in the same way, open Claude Code in a folder that holds your brief and research notes, and ask it to outline first. Its value is the loop: outline, gaps, sources, hook options, then feedback on each section as you write it. Don't skip straight to "write it".

The three that make

4. Your voice skill. I couldn't verify a published "Voice DNA" skill from an official or traceable source, so don't install one by that name from an unknown repository. Build your own instead. Install Anthropic's skills with /plugin marketplace add anthropics/skills, then /plugin install example-skills@anthropic-agent-skills, and use skill-creator with this prompt:

Act as an editor building a reusable writing-voice skill for me with skill-creator.
Inputs: [3-5 PIECES I WROTE AND LIKED], [WORDS AND PHRASES I NEVER USE], [WHO I WRITE FOR].
Steps: 1) study the samples and describe my sentence length, vocabulary, structure and humour, quoting short examples; 2) list the rules as do and don't pairs; 3) draft a SKILL.md with a name, a description of when to use it, the rules, and two short before-and-after examples.
Rules: base every rule on something in the samples. Leave out anything you're inferring. If I give you fewer than three samples, ask me for more before you draft.
Before answering, check that each rule points to a sample.

Paste in your best work, the phrases you hate, and who reads you.

5. ElevenLabs. Run npx skills add elevenlabs/skills, then set ELEVENLABS_API_KEY. The included setup-api-key skill walks you through getting the key. The pack covers text-to-speech, speech-to-text, sound effects, music, voice changer, voice isolator, dubbing and agents. For a studio you need text-to-speech first, and maybe sound effects. Leave the rest alone.

6. Remotion. Run npx skills add remotion-dev/skills. These skills teach Claude to scaffold a Remotion project, preview it in Remotion Studio, add captions and render. Ask for a short composition first, using your voiceover file, and render it before you ask for anything more ambitious.

How to choose

What to skip

Guardrails

Sources

All 751 AI guides · JulieMango plans from £17/mo