.agents/skills/pi/SKILL.md

name	description	user_invocable	arg_hint
pi	Interact with the raspberry pi sandbox - capture images, analyze scenes, transfer files, run commands. Use when working with the pi, camera, or servos.	true	capture | analyze | look | send <file> | get <file> | exec <cmd>

/pi

Raspberry Pi 3B+ sandbox toolkit. The pi is at 10.0.0.11 (user: pi). SSH key auth is set up - use regular ssh/scp/rsync via Bash, no need for ssh-mcp.

connection

PI="pi@10.0.0.11"
ssh $PI '<command>'
scp $PI:<remote> <local>
scp <local> $PI:<remote>
rsync -avz $PI:<remote_dir>/ <local_dir>/

hardware

• USB camera at /dev/video0
• Pan servo: GPIO 17, Tilt servo: GPIO 18, Trigger: GPIO 27
• Servos NOT currently wired up
• pigpiod must be started before servo use: sudo systemctl start pigpiod

commands

/pi capture [width] [height]

Take a photo and save locally.

ssh pi@10.0.0.11 'python3 ~/pi-sandbox/tools/capture.py 640 480' | base64 -d > ~/pi-sandbox/latest.jpg

Or for full resolution, capture on pi then scp:

ssh pi@10.0.0.11 'python3 -c "
import cv2
cap = cv2.VideoCapture(0)
cap.set(3, 640); cap.set(4, 480)
_, f = cap.read()
cap.release()
cv2.imwrite(\"/tmp/cap.jpg\", f)
"'
scp pi@10.0.0.11:/tmp/cap.jpg ~/pi-sandbox/latest.jpg

Then use Read tool on ~/pi-sandbox/latest.jpg to view it.

/pi analyze

Quick scene analysis - brightness, contrast, edge density, saturation.

ssh pi@10.0.0.11 'python3 ~/pi-sandbox/tools/analyze.py 640'

/pi look

Capture + analyze + view. Combines capture and analyze.

/pi send <local_file> <remote_dest>

Send a file TO the pi:

scp <local_file> pi@10.0.0.11:<remote_dest>

Or for directories:

rsync -avz <local_dir>/ pi@10.0.0.11:<remote_dir>/

/pi get <remote_file> [local_dest]

Get a file FROM the pi:

scp pi@10.0.0.11:<remote_file> <local_dest>

/pi exec

Run something on the pi:

ssh pi@10.0.0.11 '<command>'

helper scripts on pi

• ~/pi-sandbox/tools/capture.py [w] [h] - capture photo, output base64
• ~/pi-sandbox/tools/analyze.py [w] - capture + opencv stats (text output)
• ~/pi-sandbox/sentry.py - motion detection, saves to ~/pi-sandbox/captures/

constraints

• camera runs ~4fps at 640x480 on the pi 3b+
• opencv 3.2 (uses 3-value findContours return)
• tflite-runtime 2.5.0 available

project files

• pi: ~/pi-sandbox/ - main workspace
• pi: ~/pi-sandbox/tools/ - helper scripts
• pi: ~/pi-sandbox/captures/ - motion detection captures
• pi: ~/cat-deterrent/ - old cat deterrent code (has detector.py, servo.py)
• local: ~/pi-sandbox/ - local workspace + received files