Skip to content
Back to search

creator-intel

MarketplaceWorks with Claude CodeGitHub

A creator teardown from a handle: metrics, outliers, hooks, CTA and what to copy.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/mavrin-xv--creator-intel/badge.svg)](/listings/mavrin-xv--creator-intel)

HTML

<a href="/listings/mavrin-xv--creator-intel"><img src="/api/listings/mavrin-xv--creator-intel/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.

S=skills
python3 $S/creator-teardown/scripts/fetch_posts.py https://www.youtube.com/@somecreator/shorts --limit 30 -o posts.json
python3 $S/creator-teardown/scripts/profile_stats.py posts.json
python3 $S/outlier-finder/scripts/outliers.py posts.json
python3 $S/transcript-intel/scripts/transcribe.py https://www.youtube.com/shorts/VIDEO_ID --model base -o work/v1
python3 $S/transcript-intel/scripts/hooks.py work/v1.txt

Example prompt: `/teardown @juniperandrye on YouTube, I teach gluten-free baking`

Sample output, from the invented run in [examples/juniper-rye-teardown](examples/juniper-rye-teardown/teardown.md):

OUTLIERS  5.0x BREAKOUT  short  "Nobody tells you about steam"
          4.75x outlier  long   "I baked the same loaf 10 ways so you don't have to"
HOOK      "Nobody tells you this about steam. Your oven dumps it in the first
          ninety seconds."  [provocation]
CTA       spoken at 19.3 s: "Comment LOAF and I'll send it."  (comment-gate, free PDF)
COMMENTS  top question: "Where do you even buy lava rocks?" (412 likes, no video yet)
COPY      "nobody tells you" + side by side in the first frame

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.

.claude-plugin/marketplace.json · Manifest · 15 lines
  1. {
  2. "name": "creator-intel",
  3. "plugins": [
  4. {
  5. "name": "creator-intel",
  6. "source": "./",
  7. "description": "Creator and competitor research from a handle or URL: profile metrics, outlier posts, local transcripts, hook and structure breakdown, comment mining with a quote bank, 30-day trend scans and sourced competitor evidence packs. Free data path with yt-dlp and Whisper; paid scrapers optional.",
  8. "version": "0.1.0"
  9. }
  10. ],
  11. "owner": {
  12. "name": "Artem Mavrin",
  13. "url": "https://github.com/mavrin-xv"
  14. }
  15. }
README.md · README · 150 lines
  1. ![creator-intel](assets/banner.png)
  2. # creator-intel
  3. **Drop a creator's handle or URL and get a teardown: metrics, outlier posts, transcripts, hooks, CTA, audience language and what to copy, plus 30-day trend scans and sourced competitor packs.**
  4. Scrolling a competitor's feed for an hour gives you a feeling, not a reason their posts work.
  5. [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
  6. ![Works with](https://img.shields.io/badge/works%20with-Claude%20Code%20%7C%20Codex%20%7C%20any%20SKILL.md%20agent-black)
  7. ![Python](https://img.shields.io/badge/helpers-Python%203.8%2B-green)
  8. ![Data](https://img.shields.io/badge/data-yt--dlp%20%2B%20Whisper%2C%20no%20keys-orange)
  9. ```bash
  10. git clone https://github.com/mavrin-xv/creator-intel && cd creator-intel
  11. pip install yt-dlp faster-whisper # free data path; ffmpeg must be on PATH
  12. ./install.sh # or: ./install.sh --project
  13. ```
  14. <!-- install-notes:start -->
  15. Other ways in: as a Claude Code plugin, run `/plugin marketplace add mavrin-xv/creator-intel` and then `/plugin install creator-intel@creator-intel`. For Codex or another agent that reads `SKILL.md`, copy the folders under `skills/` into its skills directory. On Windows, run `install.sh` from Git Bash or WSL and type `py -3` where these docs say `python3`.
  16. <!-- install-notes:end -->
  17. Then in Claude Code: `/teardown https://www.youtube.com/@somecreator/shorts`
  18. ---
  19. ## What it solves
  20. | Pain | What the pack does |
  21. |---|---|
  22. | Top posts by raw views just show the channel's size | Outlier score: each post against the median of the 9 posts before it, Shorts and long-form apart |
  23. | Watching 30 videos to find the hook pattern | Local transcripts with timestamps, hooks cut from the first 3 seconds and labelled by type |
  24. | "They grow because they're good" | Five-dimension teardown: snapshot, engagement, hooks, structure, strategy, with evidence on each line |
  25. | Missing the comment-to-get-it funnel | Flags posts where comments run above 1.5x likes and finds the trigger word |
  26. | Guessing what the audience wants | Comment buckets (requests, questions, objections, pains) and a verbatim quote bank |
  27. | One viral post mistaken for a trend | 30-day scan scored within each source, with extra weight only when 3 or more sources agree |
  28. | Competitor decks full of opinions | Evidence pack where every bullet has a URL, a quote or an UNVERIFIED tag, checked by a script and diffed month to month |
  29. | Paying for a scraper before knowing you need one | yt-dlp and Whisper by default; any scraper API plugs in through an env var |
  30. ## What's inside
  31. | Skill | What it does | Say |
  32. |---|---|---|
  33. | `creator-teardown` | Entry point: fetch, snapshot, outliers, transcripts, comments, money path, report | "analyze @handle", "tear down this creator", "what's their strategy" |
  34. | `outlier-finder` | Posts that beat the creator's own baseline, the patterns they share, packaging ideas | "which videos overperformed", "find their outliers", "proven video ideas" |
  35. | `transcript-intel` | Local transcript, hook type, body structure, claims and CTA with timestamps | "transcribe this", "break down this hook", "why does this video work" |
  36. | `comment-mining` | Comment buckets, repeated phrases, comment-gate word, quote bank | "mine the comments", "what does the audience want" |
  37. | `trend-scan` | 30-day multi-source scan, grouped into trends, ranked by impact x confidence | "what's trending in X", "last 30 days in X" |
  38. | `competitor-research` | Sourced evidence pack: positioning, offer, ads, reviews, gaps; watch mode | "research my competitor", "find gaps", "watch this competitor" |
  39. Slash commands: `/teardown`, `/outliers`, `/transcribe`, `/comments`, `/trends`, `/competitor`.
  40. ## Quick start
  41. ```bash
  42. S=skills
  43. python3 $S/creator-teardown/scripts/fetch_posts.py https://www.youtube.com/@somecreator/shorts --limit 30 -o posts.json
  44. python3 $S/creator-teardown/scripts/profile_stats.py posts.json
  45. python3 $S/outlier-finder/scripts/outliers.py posts.json
  46. python3 $S/transcript-intel/scripts/transcribe.py https://www.youtube.com/shorts/VIDEO_ID --model base -o work/v1
  47. python3 $S/transcript-intel/scripts/hooks.py work/v1.txt
  48. ```
  49. Example prompt: `/teardown @juniperandrye on YouTube, I teach gluten-free baking`
  50. Sample output, from the invented run in [examples/juniper-rye-teardown](examples/juniper-rye-teardown/teardown.md):
  51. ```
  52. OUTLIERS 5.0x BREAKOUT short "Nobody tells you about steam"
  53. 4.75x outlier long "I baked the same loaf 10 ways so you don't have to"
  54. HOOK "Nobody tells you this about steam. Your oven dumps it in the first
  55. ninety seconds." [provocation]
  56. CTA spoken at 19.3 s: "Comment LOAF and I'll send it." (comment-gate, free PDF)
  57. COMMENTS top question: "Where do you even buy lava rocks?" (412 likes, no video yet)
  58. COPY "nobody tells you" + side by side in the first frame
  59. ```
  60. ## How it works
  61. ```
  62. handle / URL
  63. |
  64. v
  65. fetch_posts.py ---- provider: ytdlp (default) | file (any export) | http (your API)
  66. |
  67. v posts.json (one schema for every platform)
  68. |
  69. +--> profile_stats.py ...... cadence, medians, engagement, comment-gate flags
  70. +--> outliers.py ........... score vs previous 9 in the lane -> which posts to study
  71. |
  72. v
  73. transcribe.py (yt-dlp + ffmpeg + Whisper, local) -> hooks.py
  74. yt-dlp --write-comments -> mine_comments.py -> quote-bank.yaml
  75. |
  76. v
  77. teardown report (templates/teardown.md)
  78. trend items CSV -> trend_score.py -> ranked trends
  79. evidence pack.md -> evidence_check.py -> PASS / FAIL + diff vs last month
  80. ```
  81. Scripts do the counting and scoring. The agent reads the posts and transcripts and writes the judgement.
  82. ## Configure
  83. | Variable | Used by | What it does |
  84. |---|---|---|
  85. | `CI_HTTP_URL` | `fetch_posts.py --provider http` | Your scraper endpoint, with `{handle}` and `{limit}` placeholders |
  86. | `CI_HTTP_HEADERS` | same | JSON headers, for example your API key |
  87. | `WHISPER_DEVICE` | `transcribe.py` | `cpu` (default) or `cuda` |
  88. | `YOUTUBE_API_KEY` | optional, by hand | YouTube Data API for large channel histories |
  89. Instagram profiles usually need `--cookies <browser>` (your own logged-in browser) or an export from a scraper with `--provider file`. Details: [data sources](skills/creator-teardown/references/data-sources.md).
  90. ## Honest limits
  91. - Hook and comment labels are regex first passes, in English. The agent corrects them by reading; other languages need hand bucketing.
  92. - Public data has no watch time, retention or conversion. The teardown lists these as open questions instead of guessing.
  93. - yt-dlp extractors break when platforms change. Update it often (`pip install -U yt-dlp`).
  94. - Whisper can invent words over music or silence. Lines without a visible speaker are worth a check.
  95. - Trend grouping by shared title words is rough. Read the groups.
  96. - Tested live on public YouTube Shorts (fetch, transcripts, comments). Instagram, TikTok and the http provider are tested only on saved exports.
  97. ## Credits
  98. - Skill set shape (outliers, transcript intelligence, comment mining, teardown) inspired by https://github.com/ScrapeCreators/social-media-research-skills (MIT). No code copied.
  99. - 30-day multi-source scan and "several sources agree ranks highest" inspired by https://github.com/mvanhorn/last30days-skill (MIT, Matt Van Horn). Not bundled; install it for its collectors.
  100. - Evidence rule (a URL, a quote or an explicit tag on every claim) adapted from https://github.com/boringmarketer/boring-marketing-skills (MIT).
  101. - Packed transcript lines that break on pauses adapted from https://github.com/moritzkremb/claude-eyes (MIT).
  102. - Word-timestamp transcript handling also informed by https://github.com/browser-use/video-use (MIT).
  103. - Outlier score definition follows the public explanation at https://support.vidiq.com/en/articles/9660010-outliers.
  104. - Runs on https://github.com/yt-dlp/yt-dlp (Unlicense).
  105. - Transcribes with https://github.com/SYSTRAN/faster-whisper (MIT).
  106. - Or with https://github.com/openai/whisper (MIT).
  107. - Carousel slides via https://github.com/instaloader/instaloader (MIT).
  108. ## Contributing
  109. See [CONTRIBUTING.md](CONTRIBUTING.md). Run the commands in [AGENTS.md](AGENTS.md) before a pull request.
  110. <!-- author:start -->
  111. ## Author
  112. Built and maintained by [Artem Mavrin](https://github.com/mavrin-xv). This repo is one of a set of agent skill packs for marketing and operations work. Each one installs on its own; the full list is in [agent-skill-packs](https://github.com/mavrin-xv/agent-skill-packs). If a pack saved you an afternoon, a star helps other people find it.
  113. <!-- author:end -->
  114. ## License
  115. MIT. See [LICENSE](LICENSE).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.