Bird Watcher Skill
OpenClaw skill (gamora-approved)
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
🐦 Bird Watcher
Real-time bird detection for your backyard, porch, or feeder — powered by YOLOv11 and local AI.
An OpenClaw skill that turns any Mac with a webcam into a live bird detection station. YOLO identifies birds in real-time with bounding boxes, Moondream VLM identifies the species, and you watch it all from your phone, tablet, or TV via a simple web link.
Everything runs locally on your machine. No cloud. No subscriptions. No data leaves your network.
Real detection: bird in flight captured at 64% confidence with green bounding box, HUD overlay showing bird count and fps. Forestville, CA.
Second detection: bird at 53% confidence against the treeline. Outdoor deck setup, Sonoma County.
What It Does
You point a camera at your bird feeder. Bird Watcher does the rest:
- Detects birds in real-time using YOLOv11 — green bounding boxes appear the instant a bird enters the frame, even mid-flight
- Identifies species when close enough — when a detected bird is large enough in frame (50+ pixels), the system crops the region and asks a local VLM (Moondream) for species identification. Distant or fast fly-by birds get labeled "Bird" with confidence percentage. Species ID works best with birds perched nearby (feeders, railings, branches within ~15 feet of the camera).
- Streams live video to any device on your network — phone, tablet, laptop, or AirPlay to your TV
- Saves every detection — both the original frame and the annotated version with bounding boxes, timestamped
- Logs to wildlife census — if the OpenClaw
wildlife-censusskill is installed, sightings are recorded automatically
The camera feed and YOLO detection run in separate threads. The video is always smooth at full camera fps (~30fps). YOLO processes independently at ~10-15fps on Apple Silicon. You never see lag.
How It Looks
When running, the stream shows:
- Live camera feed at native resolution (1280×720)
- Green bounding boxes on detected birds with confidence percentage
- Species label (from Moondream VLM) on each box
- HUD overlay: bird count, camera fps, YOLO fps, detection counter, last species identified
- Timestamp in the corner
Requirements
Hardware
| Component | Minimum | Recommended |
|---|---|---|
| Mac | Any Mac with webcam | Apple Silicon (M1/M2/M3/M4) |
| Camera | Built-in MacBook camera | USB webcam for dedicated outdoor setup |
| RAM | 8GB | 16GB |
| Network | Not required for local use | WiFi for streaming to other devices |
Intel Macs work but expect ~5-8fps YOLO processing instead of 10-15fps.
Software
- Python 3.10 or newer — check with
python3 --version - macOS — camera permissions require macOS Security & Privacy settings
- Moondream Station (optional) — for species identification. Without it, you still get bird detection with bounding boxes, just no species names.
What This Skill Cannot Do
- It cannot access the camera without your explicit permission. macOS requires you to grant camera access interactively — no script can bypass this. You must run a command in Terminal and click "Allow."
- It cannot stream to the internet by default. The feed is only accessible on your local network. This is intentional for privacy. See the Security section below if you want remote access.
- Species ID requires proximity. Moondream is a general-purpose VLM, not a bird-specific model. It works best when birds are close to the camera — perched on a feeder, railing, or branch within about 15 feet. Distant birds and fast fly-bys are detected with bounding boxes but labeled generically as "Bird" rather than attempting an inaccurate species guess. Common backyard birds (jays, sparrows, robins, finches, hawks) at close range get reliable species identification.
- It cannot run in the background on macOS. The camera permission is tied to the foreground Terminal process. The script must run in an open Terminal window.
Installation
Step 1: Clone the repo
git clone https://github.com/MS-707/bird-watcher-skill.git
cd bird-watcher-skill
Step 2: Install dependencies
Recommended: use a virtual environment
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
pip3 install -r requirements.txt
If you get an externally-managed-environment error on newer Python:
pip3 install --break-system-packages -r requirements.txt
⚠️ Warning:
--break-system-packagesbypasses Python's environment isolation and can cause conflicts with system packages. Use a virtual environment instead whenever possible.
Verify everything installed:
python3 -c "import cv2, flask, ultralytics, requests; print('All dependencies OK')"
Step 2.5: Install Moondream Station (optional — for species ID)
Species identification requires Moondream Station running locally. Without it, birds are still detected with bounding boxes but labeled generically as "Bird" instead of by species name.
- Visit https://moondream.ai for installation instructions
- Once installed and running, verify with:
You should see a JSON response withcurl http://localhost:2020/health"server": "moondream-station". - Bird Watcher will automatically detect Moondream on startup and enable species ID.
Step 3: Grant camera permission
This is the most important step. Run this command in Terminal (not from a script):
python3 -c "import cv2; cap = cv2.VideoCapture(0); print('Camera:', cap.isOpened()); cap.release()"
macOS will show a permission dialog. Click Allow. You only need to do this once.
If it prints Camera: True — you're good. If Camera: False — go to System Settings → Privacy & Security → Camera and enable Terminal (or Python).
Step 4: Start the stream
python3 main.py
The script will print a URL like:
🐦 Bird Watcher Live Stream v3
🔐 Stream URL: http://YOUR_LOCAL_IP:8888?token=abc123xyz
Open that URL on your phone or any device on the same WiFi. The token is generated fresh each time you start the stream — only people with the URL can view your camera feed.
Note:
bird_watcher_stream.pyis kept as a backwards-compatible wrapper that callsmain.py.
Configuration
All settings can be passed via CLI flags or environment variables:
# CLI flags
python3 main.py --port 9999 # Custom port
python3 main.py --model yolo11n.pt # Nano — fastest, less accurate
python3 main.py --model yolo11m.pt # Medium — slower, more accurate
python3 main.py --confidence 0.20 # Higher = fewer false positives
python3 main.py --persist 5 # Seconds to keep bounding box visible
python3 main.py --no-save # Don't save detection frames to disk
# Environment variables
BIRDWATCH_PORT=9999 python3 main.py
BIRDWATCH_MODEL=yolo11n.pt python3 main.py
BIRDWATCH_CONFIDENCE=0.20 python3 main.py
MOONDREAM_URL=http://192.168.1.50:2020 python3 main.py
| Environment Variable | Default | Description |
|---|---|---|
BIRDWATCH_PORT | 8888 | HTTP port for stream server |
BIRDWATCH_MODEL | yolo11s.pt | YOLO model file |
BIRDWATCH_CONFIDENCE | 0.15 | Detection confidence threshold |
BIRDWATCH_PERSIST | 3 | Seconds to keep bounding boxes visible |
BIRDWATCH_TOKEN | (r |