Guide · Agent skill
How to use the Audience Persona Builder agent skill (cheat-persona)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The cheat-persona skill is an agent skill that reads your post-review notes and the top comments on your published pieces, then groups those comments into a profile of who actually watches and engages with your account. It writes that profile to an audience.md file. This guide explains what agent skills are, how to install this one, how to trigger it with natural requests, and what to check before you rely on the output.
Audience Persona BuilderXBuilderLAB/cheat-on-content · MIT Download the skill Quick start
- 1Download the ZIP. Download the ZIP for skills/cheat-persona from the item page.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder.
- 3Or upload in the Claude apps. In the Claude apps, open Settings → Capabilities → Skills and upload the skill.
- 4Prepare your project. Make sure your cheat-on-content project has predictions/*.md files with review sections containing comment data.
- 5Ask in plain language. Say a trigger phrase such as build persona or 我的观众是谁 in your agent session.
- 6Review audience.md. Open the generated audience.md, check the confidence level and which traits are verified versus hypothesised.
What an agent skill is and what this one does
An agent skill is a folder containing a SKILL.md file that tells a compatible AI agent how to carry out a specific task. This skill, cheat-persona, is designed for the cheat-on-content project and works in Claude Code or a compatible SKILL.md agent environment.
The skill answers one question: who is watching? It pulls comment data from the predictions/*.md review sections of your project, clusters the comments into groups such as self-identification, emotional reasons for commenting, points of disagreement and language style, and saves the result to audience.md. Later content-planning steps read that file so new ideas fit your real viewers.
The profile is deliberately kept separate from the scoring system. The rubric answers how a draft is scored; the persona answers who is watching. The persona never feeds into how drafts are predicted to perform.
Install the cheat-persona skill
Option one: download the ZIP for skills/cheat-persona from the item page and unzip it into your user skills folder, ~/.claude/skills/. Option two: unzip it into a project-level .claude/skills/ folder inside the project where you want it to apply.
Option three: in the Claude apps, open Settings, go to Capabilities, then Skills, and upload the skill there.
The skill's files list contains a single file, SKILL.md. After installing, the agent should be able to find the skill and respond to its trigger phrases.
- User-wide: ~/.claude/skills/
- Per project: .claude/skills/ in the project folder
- Claude apps: Settings → Capabilities → Skills, then upload
Check the requirements before you start
You need a cheat-on-content project folder. The skill reads predictions/*.md files and their review sections, and can also use benchmark.md, rubric_notes.md and videos/*/report.md if they exist. The skill declares Bash, Read, Write, Edit, Glob and Grep tool permissions in its header, so your agent environment must allow those.
The skill needs published posts with reviewed comment data. Without that, it has nothing to cluster. If you have no review data but do have a benchmark.md, the skill can produce an unverified, aspirational profile that is clearly labelled as such.
- Claude Code or a compatible SKILL.md agent environment
- A cheat-on-content project folder with predictions/*.md review sections
- Optional: benchmark.md, rubric_notes.md, videos/*/report.md
- Tool permissions: Bash, Read, Write, Edit, Glob, Grep
Trigger the skill with natural requests
You do not need a special command. Ask the agent in plain language using one of the trigger phrases from the skill description. Phrases listed in the source include 构造受众画像, 更新 persona, 我的观众是谁, build persona, 刷新受众画像 and 看看我的受众画像.
You can also pass optional arguments. The skill's argument hint lists seed-from-benchmark, which forces a benchmark-based seed, and rebuild, which rebuilds the profile even if the data has not changed.
- "build persona"
- "更新 persona"
- "我的观众是谁"
- "看看我的受众画像"
- Arguments: seed-from-benchmark, rebuild
What the output looks like
The skill runs in phases: it collects data from review sections, assigns a confidence level based on how many reviews exist, clusters the comments, cross-checks the persona against the rubric notes, rewrites audience.md, and then reports the changes.
audience.md is rebuilt in full each time, with a version number and last-rebuilt date in the header and a single appended line in the version history. Traits backed by at least 3 comments count as verified; weaker traits are placed under hypotheses. If the persona and the rubric calibration disagree, the skill flags the conflict rather than forcing a match.
- Confidence levels run from no data to stable, based on the number of reviews
- Verified traits need at least 3 supporting comments, each with its source
- Contradictions between persona and rubric are flagged for the next review
Safety notes: scripts and network
According to the item details, this skill runs no scripts and uses no network. It works by reading and writing files in your project folder with the tools listed in its header.
Because audience.md contains results from published posts, the cheat-score-blind skill is set to refuse reading it. This keeps the scoring step blind to audience data.
- Runs scripts: no
- Uses network: no
- Reads and writes files in your project using the declared tools
Tips for better results
Run a few reviews before asking for a persona. The skill performs best with several reviewed posts; with three or more, the profile is treated as data-grounded.
Treat the output as a picture of the most active commenters, not the whole audience. Silent viewers do not leave comments and do not appear in the data. Refresh the profile after a few new reviews, since it lags behind real changes in your audience.
- Review at least three posts before relying on the profile
- Add your own assumed traits if you want, but they will be labelled user-asserted and kept out of the verified section
- Use the rebuild argument after major changes in your content
What people use it for
- Checking whether your assumed audience matches the people who actually comment on your posts.
- Refreshing a living audience profile after every few new post reviews.
- Giving content-planning steps a reference for who a new idea should speak to.
- Spotting an audience segment you did not expect, such as viewers who are not your target group.
Frequently asked questions
Is the cheat-persona skill free?
The item details list the licence as MIT. The source does not mention pricing, so check the item page for any cost details.
What do I need to run the cheat-persona skill?
You need Claude Code or a compatible SKILL.md agent environment, a cheat-on-content project folder with predictions/*.md review sections, and Bash, Read, Write, Edit, Glob and Grep tool permissions. It works best with several reviewed posts.
How do I install an agent skill like this one?
Download the ZIP from the item page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.
Does the skill run scripts or use the internet?
According to the item details, it runs no scripts and uses no network. It reads and writes files in your project folder.
Can I use the persona to score my drafts?
No. The skill is designed to stay out of prediction. The persona shapes content direction for content-planning steps, while the scoring step is kept blind and is set to refuse reading audience.md.
What if I have no published posts yet?
The skill will not invent a profile without data. It tells you a persona needs review data, and suggests running reviews or importing a benchmark. A benchmark-based profile is labelled unverified.
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-persona if you publish short-video or social content, have several posts with reviewed comments, and want an audience profile grounded in what commenters actually say. Pick something else if you have no review data, want a statistically reliable audience study, or want the persona to score drafts, since the skill is built to keep it out of scoring and comments only reflect the most active viewers.
Download the skill