reverse-niche-skill
Free TikTok and YouTube outlier finder and competitor analysis skill for Claude Code, Codex and Cursor. Finds outlier videos, flops, hooks and format fatigue.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/jutsuhq--reverse-niche-skill)HTML
<a href="/listings/jutsuhq--reverse-niche-skill"><img src="/api/listings/jutsuhq--reverse-niche-skill/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
npx skills add jutsuhq/reverse-niche-skill -a claude-code -g
Or copy `skills/reverse-niche` into your agent's skills folder, for example `~/.claude/skills/`.
Requirements: `ffmpeg`, plus `yt-dlp` or [`uv`](https://docs.astral.sh/uv/). With `uv` the scripts run `uvx yt-dlp`, so nothing else needs installing.
Supported platforms
<p> <img src="assets/logos/tiktok.svg" width="36" alt="TikTok"> <img src="assets/logos/youtube.svg" width="36" alt="YouTube"> <img src="assets/logos/youtubeshorts.svg" width="36" alt="YouTube Shorts"> </p>
| Platform | Works | Data available | |---|---|---| | TikTok profile | yes | views, likes, comments, shares, saves, length, date, sound | | YouTube channel (`/videos`) | yes | views, length, approximate date | | YouTube Shorts (`/shorts`) | yes | views only; no dates, so cadence is unknown | | Twitch VODs | yes | views, length | | Instagram, X | no | need a logged-in session |
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 185 lines
- # reverse-niche: free TikTok and YouTube outlier finder and competitor analysis skill
- 
- [](LICENSE)
- [](https://agentskills.io)
- [](https://docs.claude.com/en/docs/claude-code/skills)
- **reverse-niche** is a free, open-source agent skill that finds the outlier videos on any TikTok, YouTube or YouTube Shorts channel and explains why they worked.
- It gives every video an **outlier score**: views divided by the channel's own median. A video at 3× or more is an outlier, a video at 0.3× or less is a flop.
- Then the agent watches the outliers and the flops, finds what separates them, checks whether the format is dying, and designs a better version with episode ideas.
- It runs in **Claude Code**, **Codex**, **Cursor** and any agent that reads [Agent Skills](https://agentskills.io) (`SKILL.md`).
- Everything runs locally on `yt-dlp` and `ffmpeg`: no API key, no browser extension, no subscription.
- - [Install](#install)
- - [Usage](#usage)
- - [Example](#example)
- - [What it measures](#what-it-measures)
- - [How it works](#how-it-works)
- - [Compared with other tools](#compared-with-other-tools)
- - [FAQ](#faq)
- - [Contents](#contents)
- ## Install
- ```bash
- npx skills add jutsuhq/reverse-niche-skill -a claude-code -g
- ```
- Or copy `skills/reverse-niche` into your agent's skills folder, for example `~/.claude/skills/`.
- Requirements: `ffmpeg`, plus `yt-dlp` or [`uv`](https://docs.astral.sh/uv/). With `uv` the scripts run `uvx yt-dlp`, so nothing else needs installing.
- ### Supported platforms
- <p>
- <img src="assets/logos/tiktok.svg" width="36" alt="TikTok">
- <img src="assets/logos/youtube.svg" width="36" alt="YouTube">
- <img src="assets/logos/youtubeshorts.svg" width="36" alt="YouTube Shorts">
- <img src="assets/logos/twitch.svg" width="36" alt="Twitch">
- </p>
- | Platform | Works | Data available |
- |---|---|---|
- | TikTok profile | yes | views, likes, comments, shares, saves, length, date, sound |
- | YouTube channel (`/videos`) | yes | views, length, approximate date |
- | YouTube Shorts (`/shorts`) | yes | views only; no dates, so cadence is unknown |
- | Twitch VODs | yes | views, length |
- | Instagram, X | no | need a logged-in session |
- ## Usage
- Ask your agent in plain words:
- - "Find the outlier videos on tiktok.com/@channel and tell me why they worked."
- - "Analyse this YouTube Shorts channel and design a better version of its format."
- - "Compare these 4 TikTok accounts in my niche. Which formats work, which are dying?"
- - "Is this TikTok format still growing or is the audience tired of it?"
- The skill can also be run by hand:
- ```bash
- python3 skills/reverse-niche/scripts/scan_channel.py https://www.tiktok.com/@channel scan/ --limit 60
- skills/reverse-niche/scripts/download_videos.sh videos/ <outlier-url> <flop-url>
- ```
- ## Example
- "Analyse this TikTok channel and design something better than it." The scan prints:
- ```
- mindset_channel: 30 videos, 42,992,025 views, baseline 30 videos, median 237,250, mean 1,433,068, confidence high
- mean is 6x the median: one or two hits dominate this channel, judge the rest without them
- tiers: >=3x: 9 flops <=0.3x: 10
- cadence: 1.9 videos/day over 16 days
- fatigue: newest 10 median = 0.04x channel median ... -> tiring
- ```
- The agent then watches the top 5 and 3 flops as contact sheets and writes a report:
- numbers, format, what works, what flops, 7 weaknesses, a new design with a beat sheet, and episode ideas tied to the outliers they build on.
- The full report is in [examples/sample-report.md](skills/reverse-niche/examples/sample-report.md).
- ## What it measures
- | Metric | Formula | Why it matters |
- |---|---|---|
- | Outlier score | views ÷ channel median | compares a video with the creator's normal results, not with raw views |
- | Tiers | ≥1.5×, ≥3×, ≥5×, ≥20× | 20× and above is usually a one-time cause, not a format |
- | Flop | ≤0.3× median | the same format failing shows what the outliers did differently |
- | Clean baseline | median without videos younger than 7 days and sponsored posts | new and paid videos would skew the median |
- | Confidence | from the sample size: <5, 5–9, 10–19, 20+ settled videos | under 10 videos nothing is flagged |
- | Engagement | likes, comments, shares, saves per 1000 views | saves point to useful content, shares to humour or identity |
- | Engagement lift | engagement rate ÷ the channel's median rate | high lift with average views is an under-pushed idea worth retrying |
- | Fatigue | newest 10 median ÷ channel median, and ÷ the previous 10 | below 0.5× the audience is tiring of the format |
- | Hit rate | share of videos at ≥1× median, now vs before | a falling hit rate confirms fatigue |
- | Title patterns | median outlier score per pattern (`vs`, number, question, POV, first person, how/why, list) | which title shapes work on this channel |
- | Pace | cuts per 10 s, first cut time | one take or fast edit |
- From watching the videos the agent adds the hook in three layers (seen, read, heard), the first payoff, the structure, voice, sound, captions, CTA and risks.
- ## How it works
- 1. **Scan.** `scan_channel.py` lists the channel with yt-dlp and writes a table with the outlier score, flags and engagement for every video.
- 2. **Watch.** `download_videos.sh` downloads only the outliers and 2–3 flops and makes a 24-frame contact sheet of each. Whisper transcribes the first 15 s.
- 3. **Compare.** The agent fills the same checklist for every video and compares outliers with flops in the same format. A concept that hits twice is a recipe; a single hit is a lead.
- 4. **Design.** Every weakness of the competitor gets a fix, plus one signature detail of your own. What cannot carry over to a new channel (fame, a news moment, a peaked sound) is struck out.
- 5. **Report.** Numbers, format, what works, what flops, weaknesses, design and 5–10 episode ideas.
- ## Compared with other tools
- | | reverse-niche | Outlier browser extensions | vidIQ, TubeLab and similar | By hand |
- |---|---|---|---|---|
- | Outlier score against the channel median | yes | yes | yes | yes, with a spreadsheet |
- | Runs locally, no account | yes | in your browser | no | yes |
- | Watches the videos and explains the hook and structure | yes | no | no | yes, slowly |
- | Compares outliers with flops in the same format | yes | no | no | yes, slowly |
- | Format fatigue and hit rate | yes | no | not as a metric | no |
- | Designs a better version with episode ideas | yes | no | no | no |
- Browser extensions such as SortViral or Sort Feed are the fastest way to sort a profile you have open.
- reverse-niche starts where they stop: it explains why the outliers won and what to make next.
- ### Other agent skills: ScrapeCreators and TubeLab
- | | reverse-niche | [ScrapeCreators skills](https://github.com/ScrapeCreators/social-media-research-skills) | [TubeLab skills](https://github.com/TubeLabHQ/youtube-research-skills) |
- |---|---|---|---|
- | Platforms | TikTok, YouTube, YouTube Shorts, Twitch | TikTok, Instagram, YouTube, Reddit, X, LinkedIn, Facebook, Threads and more | YouTube |
- | Needs | `yt-dlp` and `ffmpeg` | a ScrapeCreators API key (paid credits) | a TubeLab account and its MCP server (searches cost credits) |
- | Data source | public pages, fetched on your machine | ScrapeCreators API | TubeLab database |
- | Outlier ratio against the channel baseline | yes | yes | yes |
- | Reads the video itself | contact sheets of frames, cuts, first 15 s transcript | transcripts | transcripts and thumbnails |
- | Format fatigue and hit rate | yes | not documented | not documented |
- | Scope | one job: a niche teardown and a better design | many skills: comments, ads, transcripts, competitor teardowns | many skills: niche finding, thumbnails, sponsors, monetization risk |
- Pick ScrapeCreators for Instagram, X or ad libraries, and TubeLab for searching across all of YouTube for new niches.
- Pick reverse-niche when you already know the channels and want a free, local teardown of TikTok or Shorts.
- ## FAQ
- **How do I find outlier videos on a competitor's TikTok account for free?**
- Take the channel's last 30–60 videos, compute the median views, and divide each video's views by that median. Videos at 3× or more are outliers.
- `scan_channel.py` does this from a profile URL and leaves out videos younger than a week, whose views are still growing.
- **What is an outlier score?**
- Views divided by the channel's median views. It shows how a video did compared with that creator's normal results, so a 300k video on a 20k-median channel (15×) counts for more than 2M on a 5M-median channel (0.4×).
- **How do I tell if a TikTok format is dying?**
- Compare the median of the newest 10 videos with the channel median and with the 10 before them. Below 0.5× and a falling hit rate mean the audience is tiring of it. The scan prints both as `fatigue`.
- **How do I reverse-engineer a viral TikTok?**
- Watch it next to 2–3 flops from the same channel and format. Write down the hook in the first 3 s (what is seen, read and heard), the structure, length, sound and payoff for each. The difference between them is the formula.
- [`references/video-checklist.md`](skills/reverse-niche/references/video-checklist.md) lists every field.
- **Does it work for YouTube Shorts and long-form YouTube?**
- Yes. yt-dlp lists any channel, Shorts tab or playlist. Shorts listings carry no upload dates, so cadence is unknown and fatigue follows the listing order.
- **Can it see retention, watch time or click-through rate?**
- No. Those numbers are visible only to the channel owner. The skill never estimates them for a competitor.
- **Do I need vidIQ, TubeLab or another paid account?**
- No. It uses public data through yt-dlp.
- **How is it different from ScrapeCreators or TubeLab skills?**
- Both are agent skill sets built on a paid data API: ScrapeCreators covers many social networks, TubeLab covers YouTube with its own database. reverse-niche needs no key or account, fetches public data on your machine, and watches the videos as frame sheets. See [the comparison](#other-agent-skills-scrapecreators-and-tubelab).
- **Will it copy my competitor?**
- No. It keeps what makes the genre recognisable and designs an original version. It never copies names, characters, logos or scripts.
- ## Contents
- ```
- skills/reverse-niche/
- SKILL.md the workflow the agent follows
- scripts/scan_channel.py channel → table with outlier scores + summary (yt-dlp, Python stdlib only)
- scripts/download_videos.sh chosen videos → duration, pace, 24-frame contact sheet
- references/video-checklist.md what to extract from each video
- references/hook-patterns.md hook and title taxonomy, loops, borrowing formats from other niches
- references/report-template.md the report layout
- examples/sample-report.md a full worked report
- ```
- ## License
- MIT. See [LICENSE](LICENSE).
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.