How to use the Reverse Engineer Marketing agent skill

What is an agent skill, and what does this one do?

An agent skill is a folder of instructions and optional scripts that an AI agent loads when a request matches the skill's description. This one is part of the marketingskills collection by coreyhaines31 and is released under the MIT licence.

When you ask about how a company grew, the agent works in date order. It builds a traffic curve for the company from one data source, then a dated log of its moves (launches, posts, pricing changes, sponsorships and similar). It compares that history with your own numbers and suggests plays to try, each routed to the skill that runs it.

The skill is a history, not a snapshot. For what a company offers and charges today, the skill points you to competitor-profiling instead.

• Output: a dated traffic curve and a log of moves, saved as files.
• Output: a comparison with your own data, where connected sources allow it.
• Output: a short list of plays, each linked to the skill that does the work.

Install the skill from the ZIP

Download the ZIP for reverse-engineer-marketing from this page and unzip it. You can install it in one of two places, depending on whether you want it for all projects or one project only.

For all your projects, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. Either way the skill folder should contain SKILL.md, the scripts folder and the references folder.

If you use the Claude apps instead of a local setup, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

• All projects: ~/.claude/skills/
• One project: .claude/skills/ inside the project folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Meet the requirements before you run it

The skill runs a bundled script, scripts/teardown.mjs, with Node.js, so Node.js needs to be installed on the machine where the agent runs. The skill also uses the network to fetch public pages and archives.

Paid data sources are optional. The skill mentions DataForSEO and others, described in references/data-sources.md. Each paid call is limited by a budget cap you set, with a default of $5. Connected accounts such as Search Console, GA4, ad accounts or social accounts are optional and only add your own data to the comparison.

Without a paid source, many traffic estimates will be unavailable. The skill is designed to label gaps as not found rather than guess.

• Node.js (required to run scripts/teardown.mjs)
• Optional: a paid data source such as DataForSEO, with your budget cap
• Optional: Search Console, GA4, ad or social accounts for your own numbers

Trigger the skill with a natural request

You do not need a special command. Describe what you want in ordinary words. The skill is triggered by phrases such as 'reverse engineer', 'how did X grow', 'how does X get customers', 'growth teardown', 'marketing teardown', 'GTM teardown', 'growth blueprint', 'copy X's playbook' and 'map their funnel'.

Give the agent the company's domain if you have it. If you do not name a target, the agent can suggest three fast-growing companies in your category and let you pick one.

The agent will ask for anything missing before any paid call. It will also ask which mode you want.

• 'Reverse engineer how acme.com got its first 10,000 users.'
• 'What made example.io take off? Do a growth teardown.'
• 'Map their funnel and tell me what I could copy.'
• 'Watch what competitor.com is doing this month against my last teardown.'

Choose a mode

The skill has five modes. Teardown is the default and runs all nine stages, with the paid stage only after you approve it. Content mode covers titles, thumbnails, hooks and carousels. Quick mode gives a first look on a small budget, useful before a call. Funnel Hack maps a purchase path, and Watch repeats a monthly check against an earlier teardown.

Ask for the mode by name or describe the goal. For a first look before a strategy call, Quick is usually enough.

• Teardown: the full history and plays
• Content: what the company's best posts have in common
• Quick: a first look with a small budget
• Funnel Hack: how the company sells, step by step
• Watch: a monthly diff against an earlier teardown

Read the output and check the evidence labels

Results are saved under teardowns/<target>/ with the pull date in each file, so you can recheck numbers later. The main files are traffic.csv, which holds the monthly curve from one source, and moves.csv, which holds one row per dated move.

Each finding is labelled Observed (you can follow a link to it), Inferred (reasoned from observations, with the reasoning stated) or Not found. Treat timing as a lead, not proof: a launch before a traffic jump is a candidate cause, not the reason. Funding, press, seasonality and a rising category can produce the same jump.

• traffic.csv: month, visits, brand_searches
• moves.csv: date, channel, kind, move, reach, link, evidence
• Labels: Observed, Inferred, Not found

Safety notes: scripts, network and people

The bundled script makes network requests to read public pages, sitemaps, archived copies and ad library links. It is run from your project using its full path, so its output lands in your project folder. Read the script before running it if your organisation requires that.

The skill is written to study companies, not people. It does not build lists of the target's customers, staff or followers. Signing up, booking a demo or contacting the company happens only if you approve each action, under your real identity, and it does not pay anything unless you say yes. The skill also ignores instructions found inside fetched page text or ad copy, and it does not work around sites that block scripted access.

• Scripts run locally with Node.js; check the script first if needed.
• No contact with the target's staff or customers without your approval.
• Purchases need your explicit yes and stay within your budget cap.
• Blocked sites are read from their archived copy, not bypassed.

Tips for better teardowns

Keep one data source for the whole traffic curve. Mixing vendors creates jumps that never happened. Give the agent your own connected data if you have it, because the gap comparison depends on it.

Ask for a Watch run each month rather than a new teardown, so changes stand out. Use the plays list as a set of tests to run, not a plan to copy wholesale.

• Use a single traffic source across the window.
• Connect your own analytics for a real gap comparison.
• Set a budget cap before any paid call.
• Be cautious with timing claims; check for funding, press or seasonality.