Pranav Innovates

You commented SKILLS. Here’s everything from the reel.

5 Claude skills almost nobody has installed (yet)

From @PranavInnovates (the "Comment SKILLS" reel, Nov 2026). I ran every one of these myself, on my own channel's videos, before putting it in the video. Star counts are from GitHub on 6 Nov 2026.

All five are free and open source. They work in Claude Code, and most also work in Codex and Cursor. The quickest install is the skills CLI (npx skills add OWNER/REPO), which asks which agents to set up. Restart Claude Code afterwards.

geo-sleuthfinds where a photo was taken6.3K GitHub stars

github.com/Oldcircle/geo-sleuth · MIT

It works from the landscape itself: skylines, elevation data, satellite tiles, road geometry and street view. Then it shows its evidence on a satellite map.

Terminal
npx skills add Oldcircle/geo-sleuth

Then: find where this photo was taken and attach the photo.

  • Needs Python 3.10+, uv and Chrome. The first run downloads about 2.4 GB of models.
  • It does reverse image search (Baidu and Yandex) by default, which uploads your photo. To keep the photo local, say "no reverse image search" (it then runs intake with --no-rev).
  • Only use it on your own photos. The README says the same.
  • My test: a public CC0 road photo with its metadata removed and reverse search off. From the mountain skyline, the lane markings and the roadside trees, it put the camera on State Highway 8 in New Zealand's Mackenzie Basin. The photographer's own tag is on the same highway, 3.2 km away. It took about 25 minutes.

rea-videoreverse-engineers how a video was made3.1K GitHub stars

github.com/JuneYaooo/rea-video · Apache-2.0

Shot by shot, it says whether each shot is live footage, AI video, 3D, animation, code-rendered or a screen recording. It also breaks down the captions, cuts, voice (human or text-to-speech), music and sound effects, then writes a plan to remake the video.

Terminal
npx skills add JuneYaooo/rea-video
pip install "rea-video @ git+https://github.com/JuneYaooo/rea-video"   # the CLI; needs ffmpeg and Python 3.10+

Then: break down how this video was made: video.mp4 (say "in English" for an English report).

  • With no image model configured, Claude itself reviews the frames, and this worked well in my test.
  • Its remake step can call paid generation, and it asks first.

stack-diagramone sentence in, an architecture diagram out2.0K GitHub stars

github.com/caru-ini/stack-diagram · MIT

You get real brand logos, grouped boxes and clean arrows, in light and dark, as HTML, PNG or a React component.

Terminal
npx skills add caru-ini/stack-diagram

Then: Diagram our stack: Next.js and a Hono API on Vercel, Postgres on Neon, auth with Clerk, payments with Stripe.

  • PNG export needs Playwright (pip install playwright && python3 -m playwright install chromium).
  • At the end it offers to post your diagram to a public gallery, and only posts if you say yes twice. Say no if it's private.

reanimaterebuilds any animation in code, then proves it matches1.3K GitHub stars

github.com/NiazMorshed2007/reanimate · MIT

Give it a website, a screen recording, an mp4, a gif or a Lottie file. It films the original, rebuilds it in your project, films the copy the same way and compares them frame by frame. It repeats until timing and easing match.

Terminal
npx skills add NiazMorshed2007/reanimate -g -a claude-code

Then: /reanimate the hero entrance on https://example.com, into components/Hero.tsx or /reanimate this loader, 0:03 to 0:05 of ~/Downloads/loader.mov.

  • Needs Bun 1.1+ (bun.sh) on macOS or Linux. The first run downloads Chromium and ffmpeg (about 250 MB).
  • My test (3 seconds of one of my reels) took 26 minutes and 5 rounds. Every timing landed within 0.06 s of the original.

bullshit-detectorfact-checks a video, article or tweet161 GitHub stars

github.com/SerhiiKorniienko/bullshit-detector · MIT

It extracts every claim, checks each against sources with web search, and gives each one a verdict (confirmed, plausible, misleading, false or unverifiable). Then it gives a 0-10 BS score and an HTML report card.

Terminal
npx skills add SerhiiKorniienko/bullshit-detector

Or install it as a Claude Code plugin: /plugin marketplace add SerhiiKorniienko/bullshit-detector, then /plugin install bullshit-detector@serhii-korniienko. Pick one install method, not both.

Then: is this bullshit? <YouTube, article or tweet link>.

  • Needs uv. Fetching YouTube transcripts works best from a home connection; you can also paste the text.
  • I ran it on my own Short: BS score 2/10, with 18 of 19 claims confirmed. It flagged one maker claim I'd repeated without saying so.

Safety: a skill runs commands on your machine. Read its SKILL.md and any install script before you run it. None of these needs API keys.

These skills belong to their authors (links above). I'm not affiliated with them or with Anthropic.