replay

name	replay
description	Cloud app recording skill. Detects your environment, launches your app in an isolated environment, drives it with browser automation or Computer Use, records a video, and uploads to your chosen storage provider (Cloudflare R2, Hetzner, YouTube, or local). Works standalone or with a vibe-provisioned server. Complements vibe (server setup) and ship (dev cycle).
argument-hint	[{"optional":"URL or feature description to record"}]

Replay

You are recording a live video of a running app — in the cloud or locally — and uploading it to a storage provider. Interview first, then execute in one clean pass.

Args: {{args}}

Note: The bash blocks below assume a POSIX shell. On Windows, run them via the Bash tool / Git Bash.

Phase 1: Environment Detection

Silently run all checks, then summarize in 2–3 lines before the interview.

# OS / environment
uname -a 2>/dev/null || ver

# Vibe server config
cat ~/.vibe/server 2>/dev/null && echo "VIBE_SERVER_FILE_FOUND" || true
echo "VIBE_SERVER_ENV=${VIBE_SERVER:-not_set}"

# Docker
docker info 2>/dev/null && echo "DOCKER_AVAILABLE" || echo "NO_DOCKER"

# Playwright
bunx playwright --version 2>/dev/null && echo "PLAYWRIGHT_AVAILABLE" || echo "NO_PLAYWRIGHT"

# ffmpeg / Xvfb (VNC approach)
which ffmpeg 2>/dev/null && echo "FFMPEG_AVAILABLE" || echo "NO_FFMPEG"
which Xvfb 2>/dev/null && echo "XVFB_AVAILABLE" || echo "NO_XVFB"

# Wrangler (Cloudflare R2)
which wrangler 2>/dev/null && echo "WRANGLER_AVAILABLE" || echo "NO_WRANGLER"

# AWS CLI (Hetzner Object Storage)
which aws 2>/dev/null && echo "AWSCLI_AVAILABLE" || echo "NO_AWSCLI"

# Anthropic API key (Computer Use)
[ -n "$ANTHROPIC_API_KEY" ] && echo "ANTHROPIC_KEY_SET" || echo "NO_ANTHROPIC_KEY"

# App auto-detection
cat package.json 2>/dev/null | grep -E '"(start|dev|preview)"' | head -5
ls .env .env.local 2>/dev/null && grep -ihE 'PORT|URL|HOST' .env .env.local 2>/dev/null | head -5
ls docker-compose.yml docker-compose.yaml Dockerfile 2>/dev/null

From this, determine:

• Environment type: local dev machine / headless server / cloud VM
• Vibe server: detected at <host> / not detected
• Docker: available / not available
• Recording tools ready: Playwright / ffmpeg+Xvfb / neither (needs install)
• App entry point: detected <command> on port <N> / not detected

Print a 2–3 line summary to the user before Phase 2.

Phase 2: Interview

Ask all questions in a single message. Do not begin any execution until the user confirms.

1. What to record?

• Auto-detect from this project (I'll read package.json, README, routes, .env)
• A specific URL — paste it
• Describe the app and what you want recorded

2. Recording approach?

• Browser automation (Playwright --video) — best for web apps; records clicks and navigation; simplest setup (available/needs install)
• VNC + ffmpeg — full-screen capture of any app type (web, desktop TUI, CLI); needs headless X server (available/needs install)
• Computer Use API — Claude drives a real desktop VM via Anthropic; most realistic "Cursor-like" experience; requires ANTHROPIC_API_KEY (key set/not set)

3. Where to run? (Show only the relevant options based on Phase 1)

• Vibe server at <host> (recommended — already provisioned, 24/7 uptime) — only if vibe server was detected
• Ephemeral cloud container — spin up, record, destroy (needs Docker + a server)
• Local Docker — isolated container on this machine
• Directly on this machine — no isolation

4. Storage provider?

• Cloudflare R2 — S3-compatible, global CDN, cheap. Needs CLOUDFLARE_ACCOUNT_ID, R2_BUCKET, CLOUDFLARE_API_TOKEN
• Hetzner Object Storage — S3-compatible, EU-based. Needs AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, HETZNER_ENDPOINT, HETZNER_BUCKET
• YouTube (unlisted) — easy sharing, free storage. Needs YOUTUBE_CLIENT_ID, YOUTUBE_CLIENT_SECRET
• Local file only — save to ./replay-recordings/, no upload

5. Test scenarios? (skip if {{args}} already describes what to test)

• Auto-discover — I'll read the README, routes, and package.json scripts
• Describe them now — list the steps/flows to cover
• Just hit the entry point and record what loads

Once confirmed, echo back in one sentence:

"Recording <app/URL> via <approach> on <host>, uploading to <storage>, covering: <scenarios>."

Phase 3: Setup

3A: Credentials check

Verify all required env vars for the chosen storage provider before doing any recording work. See references/storage-providers.md for the exact vars per provider.

If any are missing, print exactly which vars to set and where to get them. Do not proceed until the user confirms they are set.

3B: Install recording tools

Install or verify tools for the chosen approach. See references/recording-approaches.md for exact setup commands per approach.

If using vibe server (SSH):

ssh "$VIBE_SERVER" "which bunx || npm i -g bun; bunx playwright --version 2>/dev/null || (bun add -g playwright && bunx playwright install --with-deps chromium); which ffmpeg || apt-get install -y ffmpeg xvfb"

If using local Docker (Playwright approach):

docker pull mcr.microsoft.com/playwright:v1.44.0-jammy

If using local Docker (VNC approach):

docker pull linuxserver/webtop:ubuntu-xfce

3C: Start or connect to the app

If a live URL was provided: skip — connect directly in Phase 4.

If recording a local project:

# Detect and store start command
START_CMD=$(node -e "const p=require('./package.json'); console.log(p.scripts.dev||p.scripts.start||p.scripts.preview||'')" 2>/dev/null)
APP_PORT=$(grep -iE '^PORT=' .env .env.local 2>/dev/null | head -1 | cut -d= -f2 || echo "3000")
echo "Start: $START_CMD  Port: $APP_PORT"

# Start app in background
eval "$START_CMD" &
APP_PID=$!
echo "App PID: $APP_PID"

# Wait for it to respond
for i in $(seq 1 15); do
  curl -sf "http://localhost:$APP_PORT" > /dev/null 2>&1 && echo "App ready" && break
  sleep 1
done

Set REPLAY_URL=http://localhost:$APP_PORT for use in Phase 4.

Phase 4: Record

Generate the output filename:

TIMESTAMP=$(date +%Y%m%d-%H%M%S)
OUTPUT_FILE="replay-${TIMESTAMP}.mp4"

Print [replay] Starting recording — <scenario list> before recording begins.

Execute the recording using the chosen approach. Full commands for each approach: see references/recording-approaches.md.

After all scenarios complete:

# Stop app if we launched it
[ -n "$APP_PID" ] && kill "$APP_PID" 2>/dev/null || true

# Confirm file exists
ls -lh "$OUTPUT_FILE" && echo "[replay] Recording saved: $OUTPUT_FILE"

If $OUTPUT_FILE is missing or 0 bytes, tell the user what went wrong and ask whether to retry with a different approach before continuing.

Phase 5: Upload

Upload $OUTPUT_FILE to the chosen storage provider. Full commands for each provider: see references/storage-providers.md.

After upload, store the result in VIDEO_URL.

If local only: VIDEO_URL="./replay-recordings/$(basename "$OUTPUT_FILE")"

Print [replay] Upload complete: $VIDEO_URL

Phase 6: Report

Print a clean summary:

---

Replay complete

Video	$VIDEO_URL
Recorded on	<host/local> via <approach>
Scenarios	<list of scenarios covered>
Issues observed	<none / brief list>

---

If issues were observed (errors, blank screens, crashes, unexpected UI):

Want to fix what we found?

• Run /ship <issue description> if you have the ship skill installed
• Install it: npx --yes skills add amajorai/skills/skills/ship -a claude-code -y

If no vibe server was used and the user ran locally:

Running replay on a persistent 24/7 cloud server means you can record any branch, any time, without local setup. Install vibe to provision one:

npx --yes skills add amajorai/skills/skills/vibe -a claude-code -y

Then set VIBE_SERVER=<your-server-ip> and re-run /replay.

Completion Checklist

• Environment detected and summarized
• Recording approach chosen and tools installed
• Storage credentials verified
• App started or URL confirmed
• Scenarios defined
• Video recorded and saved to $OUTPUT_FILE
• Uploaded to $VIDEO_URL (or saved locally)
• Summary printed to chat
• Upsells offered where relevant