Teach your agent
to use screencli.

--- name: screencli-record description: | Record AI-driven browser demos using the screencli CLI tool. Use this skill when: - The user wants to record a screen demo, screencast, or browser recording with an AI agent - The user says "record a demo", "screencast", "create a video", "screen record" - The user wants to export or re-encode a screencli recording - The user mentions screencli, screencli record, or screencli export - The user wants to add gradient backgrounds, effects, or auth to a recording --- # screencli — AI-Powered Screen Recording screencli records browser sessions driven by an AI agent and produces polished videos with effects. ## When to Use Use screencli when the user needs to: - Create a demo video of a web application - Record a walkthrough or tutorial driven by AI - Export recordings for social platforms (YouTube, Twitter, LinkedIn, etc.) - Add visual polish (gradients, zoom, click highlights) to recordings ## Step-by-Step: Record a Demo ### 1. Install (if not already installed) npm i -g screencli ### 2. Configure API Key (first time only) npx screencli init Running npx screencli record for the first time auto-triggers init if not configured. ### 3. Record npx screencli record <url> -p "<prompt>" [options] Required: <url> and -p, --prompt <text> Common options: --background <name> aurora, sunset, ocean, lavender, mint, ember --padding <n> Background padding % (default: 8) --corner-radius <n> Corner radius px (default: 12) --no-shadow Disable drop shadow --login Manual login before AI takes over --auth <name> Save/load auth state --no-headless Show browser window --viewport <WxH> Viewport (default: 1920x1080) ### 4. Export (optional) npx screencli export <dir> --preset <name> Presets: youtube, twitter, instagram, tiktok, linkedin, github-gif ## Authentication Flow npx screencli record <url> -p "..." --login --auth myapp # first time npx screencli record <url> -p "..." --auth myapp # reuse ## Troubleshooting - "Missing ANTHROPIC_API_KEY" — Run npx screencli init - Browser crashes — Try --no-headless - Auth expired — Re-run with --login --auth <name> - FFmpeg errors — Install FFmpeg (brew install ffmpeg) ## Reference See: references/cli-reference.md · references/effects.md

# screencli CLI Reference ## Commands ### npx screencli init Interactive setup wizard. Prompts for Anthropic API key. Saves to ~/.screencli/config.json ### npx screencli record <url> -p "<prompt>" [options] Record an AI-driven browser demo. All flags: -p, --prompt <text> Instructions (required) --background <name> Gradient preset --padding <n> Padding % (default: 8) --corner-radius <n> Corner px (default: 12) --no-shadow Disable shadow --login Manual login first --auth <name> Named auth state --no-headless Show browser --viewport <WxH> Size (default: 1920x1080) -m, --model <model> Claude model --max-steps <n> Max steps (default: 50) --slow-mo <ms> Action delay -o, --output <dir> Output dir (default: ./recordings) -v, --verbose Debug logging ### npx screencli export <dir> --preset <name> Re-encode for a platform. Presets: youtube, twitter, instagram, tiktok, linkedin, github-gif

# Gradients & Effects Reference ## Gradient Presets Six presets rendered via FFmpeg: aurora purple / blue / green sunset orange / pink / magenta ocean deep blue / teal / cyan lavender purple / pink / soft blue mint green / teal / aqua ember red / orange / amber ## Automatic Effects - Auto-trim — removes idle time - Zoom — auto-zooms to action area (--no-zoom) - Click highlights — pulse at clicks (--no-highlight) - Cursor trail — smooth overlay (--no-cursor) ## Composition Pipeline 1. Raw recording (Playwright WebM) 2. Trim dead time 3. Effects overlay (zoom, highlights, cursor) 4. Background (gradient + corners + shadow) 5. Final output (MP4 or GIF)