Guide · Agent skill
How to use the Learn From Benchmark Accounts agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Learn From Benchmark Accounts skill (cheat-learn-from) is an agent skill for the cheat-on-content repository. It takes scripts and performance numbers from a benchmark account you admire, usually a short-video creator, and extracts the patterns that make it work. After this guide you will know how to install the skill, trigger it with a plain request, and review the notes it writes into your project folder.
Learn From Benchmark AccountsXBuilderLAB/cheat-on-content · MIT Download the skill Quick start
- 1Download and unzip. Download the ZIP from this page and unzip it so you have a cheat-learn-from folder containing SKILL.md.
- 2Place the folder. Move it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder, or upload the ZIP under Settings → Capabilities → Skills.
- 3Prepare your scripts. Collect 3 to 15 script texts from a benchmark account, ideally 5 to 10, and note each video's plays, likes, comments and shares.
- 4Make the request. Say something like 学这个账号 or learn from, then give the account name.
- 5Rate each sample. For each video, answer whether it was high, medium or low impression and say why in one sentence.
- 6Review the patterns. Read the proposed patterns and rubric signals, correct anything that looks wrong, then reply ok.
- 7Check the files. Open benchmark.md, script_patterns.md and rubric_notes.md in your channel folder to confirm the imported notes.
What the skill makes the agent do
Agent skills are folders of instructions that an AI agent loads when a request matches them. This one is built for people who have no performance history of their own yet. Without history, the tool has no real signal to work from, so the skill uses a benchmark account as a reference point.
The agent asks you for 3 to 15 samples. Each sample is a script plus its numbers (plays, likes, comments, shares). It asks you to rate each one as high, medium, or low impression and why. It then identifies patterns in hooks, structure, sentence rhythm and vocabulary, and writes them into benchmark.md, script_patterns.md and rubric_notes.md.
- Minimum of 3 samples; the recommended range is 5 to 10.
- Patterns are marked as imported and untested until you verify them on your own channel.
- Rubric signals are qualitative only. The skill does not assign numeric weights.
Install the skill from the ZIP
Download the ZIP from the skill's page on this directory and unzip it. The skill is a SKILL.md file inside a folder named cheat-learn-from.
For a personal install, place the folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/cheat-learn-from/
- Project only: .claude/skills/cheat-learn-from/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Trigger the skill with a natural request
You do not need a command. Ask in plain language. The skill responds to phrases such as 学这个账号 (learn this account), 拆这几个对标视频 (break down these benchmark videos), learn from, 导入对标账号 (import a benchmark account) and 找对标 (find a benchmark).
The skill also accepts an argument hint of the form account name, followed by an optional way (a or b) and an optional append or replace flag. If you omit these, the agent asks.
- Example: 学这个账号 and then give the account name
- Example: 拆这几个对标视频 with the scripts pasted in the chat
- Example: learn from [account name] --append to add videos to an existing benchmark
Choose how to supply scripts and data
Script source has three options. The simplest is pasting script text you have already extracted (way a, the default). Way b transcribes video files you downloaded yourself with whisper-cpp or openai-whisper plus ffmpeg. Way c skips scripts and uses only titles, numbers and your impressions, which limits how deep the pattern analysis can go.
For data, you can type the numbers by hand, or use a configured perf-data adapter such as douyin-session to fetch numbers and top comments from a video URL. Hand entry needs no extra tools.
- Zero-dependency path: paste text and type numbers by hand
- Better comment data: paste text and use an adapter
- Way b needs whisper plus ffmpeg installed
Review the output and what gets written
Before anything is saved, the agent shows you the patterns it found and the qualitative rubric signals, plus a view of likely topic mix and tone. Reply ok to save, or point out any pattern you disagree with and the agent revises it.
Files are written to your channel folder. Each sample gets its own folder under samples/ with a transcript and a meta file. Re-running with replace moves the old benchmark.md to benchmark.archived/ rather than deleting it.
- benchmark.md: account info, sample table, qualitative signals, topic direction
- script_patterns.md: a new section marked untested
- rubric_notes.md: benchmark-derived initial signals
- samples/<account>/<video-id>/: source (way b only), transcript.md, meta.md
What people use it for
- A new creator with no videos yet collects five high-performing videos from a favourite account to set a reference point before brainstorming topics.
- A marketer builds a reference set of competitor videos and compares hook types and sentence rhythm across them.
- An experienced creator with some history runs one benchmark as a sanity check before planning a new series.
- A user appends new benchmark videos over time with the append flag to keep the reference set current.
Frequently asked questions
What do I need before I can use the skill?
You need the cheat-on-content repository's state and template files, since the skill works within them. You also need your benchmark scripts and their performance numbers. Video transcription is optional and needs whisper-cpp or openai-whisper plus ffmpeg, and automatic data fetching needs a perf-data adapter such as douyin-session.
Is the Learn From Benchmark Accounts skill free and what is its licence?
The skill is released under the MIT licence, as listed on its page. The source does not state pricing for the skill itself, and third-party script tools mentioned in the guide may have their own free or paid tiers.
How do I install the cheat-learn-from skill?
Download the ZIP from the skill's page and unzip it. Place the cheat-learn-from folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder, or upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
Does the skill download videos for me?
No. The skill does not fetch videos directly. You download the videos yourself, and it can then transcribe files you placed in the samples folder. Downloading tools such as yt-dlp or BBDown are something you run on your own.
Can the skill copy a creator's scripts word for word?
No, that is not its purpose. It extracts patterns in hooks, structure, rhythm and vocabulary, and it is not meant to reproduce scripts verbatim. The skill also notes that you need rights to any third-party material you collect.
How is this different from scoring my content with numeric weights?
The skill deliberately gives qualitative direction only and does not assign numeric weights, because 5 to 10 samples would overfit. Benchmark signals are a starting reference, and your own calibration samples should drive scoring later.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you are starting a channel with no performance history, or you are a marketer who wants a reference set of inspiration videos to study. Pick something else if you only want to copy another creator's content, or if you need numeric scoring; the skill intentionally avoids both.
Download the skill