Skip to content

Guide · Agent skill

How to use the Cheat Seed Topic Chat agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

cheat-seed is an agent skill for content creators who want help choosing a topic. Instead of handing you a long list of generic ideas, it talks with you about a real experience or opinion you bring, narrows to one angle, and writes a single draft. This guide explains what an agent skill is, how to install cheat-seed, how to trigger it with natural requests, and what it needs before you start.

Cheat Seed Topic ChatXBuilderLAB/cheat-on-content · MIT

Quick start

  1. 1Download the ZIP. Get the cheat-seed ZIP from this page.
  2. 2Unzip into a skills folder. Extract it into ~/.claude/skills/ for all projects, or into .claude/skills/ inside one project.
  3. 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
  4. 4Set up the workflow files (optional). Add .cheat-state.json and the other workflow files if you want history and rubric support; otherwise run with the conversation alone.
  5. 5Start a conversation. Type /cheat-seed followed by a real experience or opinion, or use a trigger phrase such as 找选题.
  6. 6Answer one question at a time. Reply to each follow-up so the agent can narrow to one angle.
  7. 7Take the draft or the candidate. Say yes to get one draft, change the angle, or write it yourself and keep the candidate.

What an agent skill is and what cheat-seed makes the agent do

An agent skill is a folder containing a SKILL.md file. The file tells a compatible agent when to use the skill and how to behave while doing so. cheat-seed is one such skill, and its folder is skills/cheat-seed in the XBuilderLAB/cheat-on-content repository. Its licence is MIT.

By default, cheat-seed works one topic at a time. You give it a theme or a real experience, and the agent asks follow-up questions about that input: what triggered it, which detail stands out, which angle you want. Once the angle is narrowed, it proposes one angle and, if you agree, writes one draft script. The skill states that the agent does not fire three open questions at you and does not hand you five candidates at once.

An optional batch mode, run with /cheat-seed --batch 5, uses an older brainstorm flow that produces several candidates and drafts.

Install cheat-seed: download the ZIP and unzip it

Download the ZIP of the cheat-seed skill from this page. Then choose one of two locations. For a personal install available in all projects, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.

If you use the Claude apps instead of Claude Code, you can upload the skill under Settings → Capabilities → Skills.

  • Personal install: ~/.claude/skills/
  • Project install: .claude/skills/ inside your project folder
  • Claude apps: Settings → Capabilities → Skills (upload)

Trigger cheat-seed with natural requests

You do not need a special syntax for the normal flow. The skill's description lists trigger phrases: "找选题" (find a topic), "我想做一条 X" (I want to make a video about X), "最近有个想法" (I've had an idea recently), "seed", and "启动种子" (start the seed). You can also type /cheat-seed followed by your topic or experience.

The skill sorts your input into modes automatically. If you name a specific thing you experienced or felt, it goes straight into a deep dive. If you give only a broad direction, it asks one question about why you want to cover it. If you say you have no idea, it moves to an external-material mode. If you type only /cheat-seed with nothing else, it asks what you want to do today.

  • Specific experience: "/cheat-seed 我最近开会被领导..." (the skill digs into that moment)
  • Broad direction: "想做职场" (it asks why you want this topic)
  • No idea yet: "帮我想个题" (it uses external sources if available)
  • Batch: "/cheat-seed --batch 5"
/cheat-seed --batch 5

Example: a one-topic session

Suppose you write: "我想做一条关于开会被领导打断的视频" (I want a video about being interrupted by my boss in meetings). The agent does not list alternatives. It asks what moment triggered the idea, which detail made you angry or amused, and who you imagine as the audience. It asks one question at a time and stops after at most four rounds of questions in the deep-dive phase.

When it has enough, it proposes a single angle with a one-sentence thesis, a suggested structure, a hook, the main points and a closing line. It then asks whether you want a draft. Answer yes to get one draft sized to your video length, or say you will write it yourself and the candidate is saved for later.

What cheat-seed needs before it runs well

The skill can run with only SKILL.md, but its instructions expect the wider cheat-on-content workflow. It reads files such as .cheat-state.json (calibration samples, typical video duration, cadence), rubric_notes.md (the scoring rubric), script_patterns.md (structure patterns), a predictions/ folder of past posts, audience.md, and benchmark.md. If .cheat-state.json is missing, the skill tells you to run /cheat-init first.

Without these files, the skill still holds a conversation, but it cannot use your history, benchmark or rubric. This SKILL.md does not create those files.

  • .cheat-state.json (required by the workflow)
  • rubric_notes.md, script_patterns.md
  • predictions/*.md, audience.md, benchmark.md (optional context)
  • Optional helper skills referenced in the instructions: aihot, trendradar-mcp, humanizer

Safety notes: scripts and network access

According to the listing, this skill does not run scripts. It does use the network: its allowed tools include WebFetch, and the instructions can pull trending topics from aihot or trendradar-mcp when those are available, or ask you to paste links manually. The skill never pulls external data on its own during a deep dive on your topic; for news topics it asks first.

Its allowed tools also include Bash, Read, Write, Edit and Glob, so review the SKILL.md before installing if you are particular about what an agent can touch on your machine.

  • Runs scripts: no
  • Uses network: yes (optional trend lookups)
  • External data is only fetched when you say yes, or when you ask for ideas and sources are available

What people use it for

  • Turning a frustrating meeting into one focused video script
  • Choosing between critique, self-reflection and a wider lesson for an opinion you already hold
  • Getting one draft sized to your typical video length from the workflow's duration setting
  • Generating five candidate topics and drafts in batch mode when you want to plan ahead
  • Checking a topic against your audience notes before you commit to writing it

Frequently asked questions

Is cheat-seed free to use?

The skill is published under the MIT licence, according to this page. The SKILL.md itself does not mention pricing, and this guide does not claim any cost beyond the licence.

How do I install the cheat-seed skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can instead upload it under Settings → Capabilities → Skills.

What do I need to run cheat-seed?

It is designed for Claude Code or another SKILL.md-compatible agent. The full workflow expects optional files such as .cheat-state.json, rubric_notes.md and script_patterns.md; without them the skill cannot use your history or rubric.

Does cheat-seed run scripts or access the internet?

The listing says it runs no scripts and does use the network. It can fetch trending topics through referenced tools or use WebFetch when you ask for ideas, and it asks before pulling external data on news topics.

How is cheat-seed different from a normal brainstorm prompt?

By default it discusses one topic at a time and digs into your own experience instead of listing many generic ideas. Batch mode, with --batch N, is the older flow that produces several candidates and drafts.

Why won't cheat-seed verify trending topics by itself?

It relies on other skills or tools such as aihot, trendradar-mcp or manual paste, which are referenced but not included in this SKILL.md. You need those set up separately if you want live trend data.

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 cheat-seed if you have real experiences or opinions but struggle to choose one specific angle, and you want a slower, one-on-one conversation that ends in a single draft. Pick something else if you want a quick list of many ideas, or if you have no existing content workflow files and no interest in building them.

Download the skill