Guide · Agent skill
How to use the Cheat Score agent skill: single draft rubric scoring
Updated · written with AI from the project's own documentation and checked against our catalogue data
Cheat Score is an agent skill that grades a single written draft, such as a short-video or social post script, against the scoring rubric in your project and prints the results to the console. It is meant for creators and teams who want a quick quality check before running a formal prediction. This guide explains what the skill does, how to install it, how to trigger it with natural requests, and what it needs to run.
Cheat Score: Single Draft Rubric ScoringXBuilderLAB/cheat-on-content · MIT Download the skill Quick start
- 1Download and unzip. Download the cheat-score ZIP from this page and unzip it to get SKILL.md.
- 2Place the skill folder. Copy the folder into ~/.claude/skills/ for personal use or into .claude/skills/ in your project. Alternatively, upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Prepare your project. Confirm that rubric_notes.md and .cheat-state.json exist in the project root. If .cheat-state.json is missing, run cheat-init first.
- 4Write or pick a draft. Save the script or post text as a file so you can refer to its path.
- 5Ask for a score. Send a request such as 打分这篇 draft.md or score this draft.md, or run /cheat-score draft.md.
- 6Read the console output. Review each dimension's score and reason, then check the composite and the suggested next step.
What is an agent skill and what does cheat-score do?
An agent skill is a folder containing a SKILL.md file that tells an AI agent how to carry out a specific task. When a request matches the skill, the agent follows the instructions in that file.
Cheat Score (skill name cheat-score) reads your draft and your project's rubric_notes.md, scores each rubric dimension from 0 to 5 with a one-line reason, and calculates a weighted composite score. The output is printed to the console only. The skill writes no files and makes no performance prediction.
- Scores are whole integers from 0 to 5.
- Each dimension gets a short reason, which helps you see why a score was given.
- Scoring is delegated to a separate blind sub-agent that sees only the draft and the rubric, so earlier conversation or past results do not bias the grade.
- The skill does not update .cheat-state.json and does not change the rubric.
Install the cheat-score skill
Download the ZIP for cheat-score from this page and unzip it. The package contains a single file, SKILL.md. Place the skill folder in one of the following locations, depending on how you want to use it.
For a personal install available across projects, put the folder in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder. In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/ (available in every project)
- Project only: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload the skill
Set up your project before scoring
Cheat Score expects your project to already contain rubric_notes.md and .cheat-state.json. These files are created by cheat-init. If .cheat-state.json is missing, the skill tells you to run /cheat-init first and then stops.
The rubric in rubric_notes.md defines the dimensions, their weights and the composite formula. The skill only applies the rubric defined in your own project files. If the file has been restructured by hand, the skill asks which line holds the current formula rather than guessing.
- rubric_notes.md in the project root
- A .cheat-state.json in the project root, created by cheat-init
- Claude Code or a compatible agent with Read, Glob and Grep tools
- Task tool support, so the cheat-score-blind sub-agent can be started
How to trigger the skill with natural requests
You can invoke the skill with its slash command or with one of the trigger phrases defined in the skill. Supply the draft path where you can, since the skill needs it to find the file.
If you do not give a path, the skill asks for one in the conversation.
- 打分这篇 [path]
- score this [path]
- 给这稿子打分
- 先打分看看
- /cheat-score draft.md
/cheat-score draft.mdRead the output and choose a next step
With the default full output, the console shows a table with each dimension, its score and its reason, followed by the formula and the composite score. The output also names the bucket the composite falls into, based on the bucket boundaries in starter-rubrics, and suggests a next step.
If you set OUTPUT_DETAIL to compact, the reasons column is removed and you see only the score table and composite.
- Full output: scores, reasons, formula and composite
- Compact output: scores and composite only
- Suggested next steps: start a prediction when the draft is final, rescore after edits, or look for past samples with a similar composite
/cheat-score draft.md — OUTPUT_DETAIL: compactSafety notes: scripts, network and what the skill does not do
Cheat Score runs no scripts and uses no network access, according to its listing. Its allowed tools are Read, Glob and Grep. It does not write any files, including predictions, rubric_notes.md or candidates.md.
The skill will not give bucket probabilities or forecast whether a post will go viral. For predictions, use cheat-predict. The skill also will not skip the blind sub-agent, because scoring without isolation defeats its purpose.
- No scripts are run and no network use is declared
- No files are written or modified
- No bucket probabilities or virality forecasts
- No rubric changes are proposed, even if the scores look unusual
Tips for getting useful scores
Score a draft before you commit to a prediction, so you can compare versions cheaply. Because scoring leaves no record, you can score as many times as you like while editing.
Keep your draft text in a single file so the path is simple to pass. If you score the same draft three or more times in a row, the skill suggests that you may be ready to decide.
- Use the reasons as diagnostics: if a reason does not match your reading of the draft, note which dimension to review
- Keep rubric_notes.md in its expected structure so the formula can be parsed
- Use cheat-predict once you have a final version
What people use it for
- Checking a short-video script against the project rubric before committing to a formal prediction
- Comparing several versions of the same script by their composite scores
- Getting a quick, non-recorded quality read while editing a draft
- Reviewing per-dimension reasons to see which part of a draft a rubric dimension is judging weakly
Frequently asked questions
How do I install the cheat-score skill?
Download the ZIP from this page and unzip it to get SKILL.md. Place the folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
What are the requirements for cheat-score?
You need Claude Code or a compatible agent with Read, Glob and Grep tools, plus Task tool support so the blind sub-agent can run. Your project must also contain rubric_notes.md and .cheat-state.json, which are created by cheat-init.
Is cheat-score free to use?
The source lists the item's licence as MIT. The source does not state any price, so check the directory page for current terms.
Does cheat-score predict whether my post will go viral?
No. The skill explicitly makes no performance prediction and refuses bucket probabilities. It only calculates the composite from your project's current rubric. Use cheat-predict for a formal prediction.
Why does the skill say it cannot find .cheat-state.json?
The skill needs .cheat-state.json in the project root to read the current rubric version and mode. If it is missing, run /cheat-init first and then try again.
How does cheat-score compare with cheat-predict?
Cheat Score is a lightweight exploration step that writes nothing and does not predict. Cheat-predict is the formal step that requires a blind check and writes an immutable log. Score is for trying versions; predict is for committing to one.
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
Cheat Score suits creators and teams who already use the cheat-on-content workflow and want a quick, unrecorded rubric check on a draft. It is not suited to predicting virality or to scoring with a generic writing rubric, since it only applies the rubric in your own project files.
Download the skill