Guide · Agent skill
How to use the Blind Prediction Logger agent skill (cheat-predict)
Updated · written with AI from the project's own documentation and checked against our catalogue data
cheat-predict is an agent skill from the cheat-on-content project that records a prediction of how a finished video script will perform before it is published. It is for creators who want to compare predicted and actual results later, using the cheat-on-content files. This guide explains what the skill does, how to install it, how to trigger it, and what it needs before it runs.
Blind Prediction LoggerXBuilderLAB/cheat-on-content · MIT Download the skill Quick start
- 1Download and unzip. Download the skill ZIP from this page and unzip it into ~/.claude/skills/ or the project's .claude/skills/ folder.
- 2Prepare the project. Make sure rubric_notes.md, .cheat-state.json, scripts/ and predictions/ exist, with the companion cheat-on-content files available.
- 3Write your final script. Save the final script in scripts/ using the scripts/<date>_<id>_<short>.md naming pattern.
- 4Ask for the prediction. Say start prediction followed by the script path, or use one of the other trigger phrases.
- 5Check you have not seen results. Confirm you have not viewed post-publication data, because the skill refuses predictions once results have been seen.
- 6Review the draft. Read the scores, bucket, distribution and scenarios, then reply ok or name the part to change.
What an agent skill is and what cheat-predict does
An agent skill is a folder containing a SKILL.md file that tells an AI agent how to carry out a specific task. cheat-predict is one such skill, and its source file is SKILL.md.
When you ask for a prediction, the agent reads your script and the rubric_notes.md scoring rubric, then has a separate sub-agent (cheat-score-blind) score the script without outside context. The main agent then drafts a probability distribution and counterfactual scenarios for you to review. Once you approve, the prediction is saved as a file that is meant to be locked against edits.
The skill refuses to make a blind prediction if you have already seen post-publication data such as views or likes. It also detects an existing prediction for the same script and appends a v2 section instead of overwriting it.
- Reads the final script and the scoring rubric
- Scores through a sub-agent that works without your conversation history
- Drafts bucket, probability distribution and scenarios for your review
- Saves the approved prediction to predictions/, with v2 appended when a v1 section already exists
Install cheat-predict from the ZIP
Download the ZIP for the skill from this directory page, then unzip it. You can install it for yourself across all projects, or only for one project.
For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. Alternatively, in the Claude apps you can upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Single project: .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills, then upload
Set up the cheat-on-content project files
cheat-predict is built for the cheat-on-content workflow and expects its project files. Before you run it, make sure your project has rubric_notes.md at the root, a .cheat-state.json state file, and scripts/ and predictions/ folders.
The source also lists companion skills and shared-references files from the cheat-on-content repository. Without them, the skill's links to its protocol and state rules will not resolve.
- rubric_notes.md in the project root
- A .cheat-state.json state file
- scripts/ for your drafts and predictions/ for saved predictions
- Companion skills and shared-references files from the cheat-on-content repository
Trigger the skill with a natural request
Name the script you want to predict, for example by path. The source lists these trigger phrases: 启动预测, start prediction, 给这稿子打分并预测, and 写预测日志.
The skill expects a script path in the form scripts/<date>_<id>_<short>.md. If you give a file outside scripts/, it will warn you and offer to copy it into place. Optional arguments include --skip-blind, which runs the scoring in the main agent instead of the sub-agent.
- start prediction scripts/<date>_<id>_<short>.md
- 给这稿子打分并预测 (score this script and predict)
- 写预测日志 (write a prediction log)
- Optional: add --skip-blind to score without the sub-agent (not recommended)
start prediction scripts/<date>_<id>_<short>.mdReview and approve the draft
Before anything is saved, the agent shows you the full draft. If you agree, reply ok and the prediction is written to disk. If a dimension or judgement looks wrong, say which one and what it should be; the agent revises and shows the draft again.
If the sub-agent's score differs from the main agent's estimate by 2 or more on any dimension, you are asked to choose which score to use.
- Reply ok to accept the draft and save it
- Name a dimension and the value you think is right to revise it
- When sub-agent and main estimates differ by 2 or more, choose a, b or c
Safety notes: scripts and network
The skill's allowed tools include Bash, Read, Write, Edit, Glob and Task. The source says the skill runs scripts, uses Bash for JSON validation and hashing with Python 3, and does not use the network.
Review the SKILL.md before installing, and note that it can write files in your project. Keep your rubric and state files in version control so changes are visible.
- Runs scripts: yes (Bash, with Python 3 for JSON validation and hashing)
- Uses network: no, according to the source
- Can read and write files in the project, including predictions/
What people use it for
- Logging a blind prediction for a script before you film and publish it, to compare against real results later
- Writing a v2 prediction after you revise a script that already has a v1 prediction
- Checking how the sub-agent's blind scores compare with your own estimate for a script
- Building a calibration history across several published videos in the cheat-on-content workflow
Frequently asked questions
Is cheat-predict free to use?
The source gives the licence as MIT. It does not state any price, so the MIT licence is the only licence information available here.
What do I need to run cheat-predict?
You need the cheat-on-content project files, including rubric_notes.md, .cheat-state.json, and scripts/ and predictions/ folders. You also need Claude Code with the Task tool, Bash access and Python 3, plus the companion skills and shared references from the repository.
How do I install cheat-predict?
Download the ZIP from the directory page and unzip it into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.
Why does cheat-predict refuse to predict?
It refuses if you have already seen post-publication data such as views, likes or comments, because then the prediction is not blind. In that case the source suggests recording a reconstructed retrospective instead.
What happens if I already have a prediction for the same script?
The skill detects an existing ## 预测 or ## 预测 v1 section and appends a ## 预测 v2 section rather than overwriting it. It also warns you if the script has not changed since the earlier prediction.
Can I use it without the sub-agent?
Yes, with the --skip-blind option, which makes the main agent score the script itself. The source says this is for cases such as when the Task tool is unavailable, and that the skill will keep warning that the prediction was self-scored.
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-predict suits creators who already use the cheat-on-content workflow and want an honest record of predicted versus actual video performance. It is not suited to anyone expecting an accurate forecast, since the output is an AI judgement limited by the rubric and calibration data, or to anyone who has already seen the results.
Download the skill