The Peekaboo content skill on GitHub

Run an open-source agent skill that turns your measured AI visibility gaps into a publish-ready draft.

3 minute read · Last reviewed: August 14, 2026

Peekaboo can tell you exactly which prompts you are missing from and which domains get cited instead. Turning that into published pages is the slow part, and it usually stalls between the dashboard and the writer. The Peekaboo content skill is a free, open-source agent skill that carries a gap all the way through to a finished draft, using your own API data as the evidence.

What it is

The skill lives in a public GitHub repository, aipeekaboo-content-skill. Point an AI agent at it, give it your Brand ID, and it connects to the AI Peekaboo API, maps where your brand is absent from AI-generated answers, and writes content aimed at being cited by ChatGPT, Perplexity and Google AI Overview.

It is a skill rather than a hosted feature, so it runs inside your own agent, on your own machine, with your own API key.

What you need

Requirement	Detail
Plan	Any active paid plan carries API access. Grow or higher in practice: Starter and Peek keys allow 5 calls a minute and 50 a day, and a single skill run exhausts that
API key	A Read key, generated from `Settings → Integrations`
Brand ID	The id of the brand you want analyzed
Agent	Claude Code, or any agent you can paste a system prompt into

The REST API article covers generating and scoping the key, including why the full key is only shown once. A read-only key is all the skill needs. The API's /brands endpoint returns the id for every brand in your project if you are not sure which one to use.

Running it in Claude Code

git clone https://github.com/filipelinsduarte/aipeekaboo-content-skill.git
cd aipeekaboo-content-skill
claude

Then ask it to run the skill for your brand:

Run the AI Peekaboo content skill for brand ID YOUR_BRAND_ID

Claude Code picks up CLAUDE.md and SKILL.md from the repository automatically, so there is nothing else to configure.

Running it in another agent

ChatGPT, Cursor and Windsurf work the same way with one extra step: copy the contents of prompt.md from the repository into your agent's system prompt or context window, then give it the same instruction with your Brand ID.

How a run works

The skill moves through six steps, and it pauses for your approval before it writes anything.

Pull the data. It reads your snapshot from the API: prompt scores, cited domains and the content types those citations belong to.
Build the gap matrix. Every tracked prompt is classified as Not Found, Rarely Shown, Sometimes Shown or Consistently Shown, so the size of the problem is explicit.
Read the citation patterns. It looks at which domains and which content formats the models actually cite in your category, rather than assuming a blog post is the answer.
Score the recommendations. Gaps are ranked by commercial intent and how contested they are, and the top three surface with a business case attached.
Scaffold the piece. Before drafting, it lays out the structure section by section: headings, word counts, calls to action and which citations belong where.
Write the draft. It produces the finished article in a citation-friendly shape, with answer-first openings, question-shaped headings, definition blocks, comparison tables and an FAQ section.

What you get back

A plain-language visibility snapshot, with the metrics explained rather than dumped
A gap matrix table covering every tracked prompt and its status
A citation pattern analysis naming the top domains and formats the models prefer
Three prioritized content recommendations, each with the business case for doing it
A content scaffold you confirm before any drafting starts
A finished draft article, matched to your brand voice or a default B2B style

The examples/ folder in the repository holds a full sample run end to end, which is the fastest way to see the output shape before pointing it at your own account.