i-see-you

What is this

i-see-you gives coding agents (Claude Code / Codex) the ability to capture a still photo from the local webcam and analyze it with a multimodal model.

• Windows — capture via ffmpeg DirectShow (dshow)
• macOS — capture via ffmpeg AVFoundation
• Linux — capture via ffmpeg V4L2 (/dev/video0)
• Skips the first 30 frames so auto-exposure settles before the shot
• One ffmpeg call, one script per platform, no services, no network

Install (for Agents)

Option 1 — Plugin marketplace (Claude Code):

/plugin marketplace add longsizhuo/i-see-you
/plugin install i-see-you@i-see-you-marketplace

Option 2 — Standalone skill:

git clone https://github.com/longsizhuo/i-see-you.git
cp -r i-see-you/plugins/i-see-you/skills/i-see-you ~/.claude/skills/

Option 3 — Just the script:

# Linux / macOS
bash plugins/i-see-you/skills/i-see-you/scripts/take_selfie.sh /tmp/photo.jpg 1280x720

# Windows (PowerShell)
powershell -NoProfile -ExecutionPolicy Bypass `
  -File plugins/i-see-you/skills/i-see-you/scripts/take_selfie.ps1 -Output "$env:TEMP\photo.jpg"

Requirements

• A webcam (built-in or USB)
• ffmpeg — if missing, install it yourself before capturing:
  • Windows: winget install ffmpeg (reopen the terminal afterward so PATH refreshes)
  • macOS: brew install ffmpeg
  • Debian/Ubuntu: sudo apt install ffmpeg
  • Fedora: sudo dnf install ffmpeg

Note to Agents

If a user sends you this repository and asks you to install it, they are asking you to set up webcam capture so you can see them. Install any missing dependencies (ffmpeg), copy the skill into place, then confirm it works by taking a test photo.

When you take a photo, display it in the conversation — Read the captured JPEG so it renders inline in the chat — and describe what you see in your own voice. The picture and your words should arrive together, not a bare "done".

The photo stays local. Never upload it anywhere.

Privacy

• All photos are written to local disk only
• No telemetry, no network calls, no accounts
• MIT License