/macos-screenshot — Capture macOS screen

Capture screenshots from remote macOS nodes and send them to users. Avoids context bloat by using proper file-based workflows instead of base64 encoding.

⚠️ Anti-Pattern: Never Do This

# WRONG: Base64 encoding a screenshot will blow up your context
nodes.run: base64 -i /tmp/screenshot.png  # 200K+ tokens = instant failure

Correct Workflow

Step 1: Capture Screen

Use screen_record with minimal duration (1ms) to capture a single frame:

{
  "action": "screen_record",
  "node": "<node-id>",
  "durationMs": 1
}

Returns: FILE:/tmp/openclaw-screen-record-<uuid>.mp4

Step 2: Extract Frame

Use ffmpeg on the gateway host to extract a PNG frame:

ffmpeg -i /tmp/openclaw-screen-record-<uuid>.mp4 \
  -frames:v 1 -update 1 -y /tmp/screenshot.png 2>/dev/null

Step 3: Send Image

Send the image file directly via the message tool:

{
  "action": "send",
  "channel": "telegram",
  "target": "<chat-id>",
  "media": "/tmp/screenshot.png"
}

Complete Example

# 1. Capture
result = nodes(action="screen_record", node="mac-xxx", durationMs=1)
# Returns: FILE:/tmp/openclaw-screen-record-abc123.mp4

# 2. Extract frame
exec("ffmpeg -i /tmp/openclaw-screen-record-abc123.mp4 -frames:v 1 -update 1 -y /tmp/mac-ss.png 2>/dev/null")

# 3. Send to user
message(action="send", channel="telegram", target="123456", media="/tmp/mac-ss.png")

Analyzing Screenshots

To analyze what's on screen, use the image tool after extraction:

{
  "image": "/tmp/screenshot.png",
  "prompt": "Describe what's visible on this macOS screen"
}

Node Discovery

Find available macOS nodes:

{
  "action": "status"
}

Look for nodes with platform: "macos" and connected: true.

Troubleshooting

Key Points

• Never base64 encode images into the conversation
• Always use file paths with the media parameter
• Screen recording permission required on macOS node
• Clean up temp files periodically to save disk space