| name | simulator-automation |
| description | Automate the EvenHub glasses simulator via its HTTP API. Use when testing or controlling the simulator programmatically — sending glasses input (up, down, click, double click), capturing screenshots, or reading browser console logs. |
| allowed-tools | ["Read","Grep","Glob","Bash","Write","Edit"] |
| argument-hint | ["automation task description"] |
EvenHub Simulator Automation
The simulator exposes an HTTP API on localhost when launched with --automation-port <PORT>.
evenhub-simulator https://my-app.example.com --automation-port 9898
Base URL: http://127.0.0.1:<PORT>
API Reference
Health check
GET /api/ping
→ "pong"
Glasses screenshot
Returns the current LVGL framebuffer as image/png (576×288, RGBA).
GET /api/screenshot/glasses
→ image/png binary (RGBA)
RGBA format — critical for image analysis:
- Background pixels:
(R=0, G=255, B=0, A=0) — transparent (alpha = 0)
- Lit/text pixels:
(R=0, G=255, B=0, A=255) — opaque (alpha = 255)
Always work in RGBA mode. Do NOT convert to RGB — it drops the alpha channel and makes
background pixels indistinguishable from text pixels (both appear as pure green).
from PIL import Image
img = Image.open('glasses.png')
pixels = img.load()
is_lit = lambda px: px[3] > 0
Webview screenshot
Captures the main browser webview using html2canvas. Returns image/png. May take a few seconds; times out after 10s.
GET /api/screenshot/webview
→ image/png binary
Console logs
Returns captured console.* output, uncaught exceptions, unhandled promise rejections, and failed fetch calls from the main webview.
GET /api/console
→ { "entries": [...], "total": N }
Each entry:
{ "id": 0, "level": "log|warn|error|info|debug|trace", "message": "...", "ts": 1712150400000 }
Prefixes for non-console sources:
[uncaught] ... — uncaught exception
[unhandledrejection] ... — unhandled promise rejection
[fetch] ... — failed fetch (non-ok status or network error)
Poll for new entries only:
GET /api/console?since_id=42
→ entries with id > 42
since_id must be a non-negative integer. Passing a negative value (e.g. -1) causes
the server to return an error, not an empty list. For the first poll, omit the parameter
entirely to retrieve all current entries, then track last_id from the response:
resp = requests.get(f"{BASE_URL}/api/console")
entries = resp.json()["entries"]
last_id = max((e["id"] for e in entries), default=0)
resp = requests.get(f"{BASE_URL}/api/console?since_id={last_id}")
Clear the buffer:
DELETE /api/console
Timing caution: Startup logs (e.g. a "ready" signal) are emitted once and lost if
you clear the buffer before reading them. Pattern: poll for the ready signal first,
then clear if needed.
poll_until(target="APP_READY")
requests.delete(f"{BASE_URL}/api/console")
Glasses input
Send a touchpad action to the glasses display.
POST /api/input
Content-Type: application/json
{ "action": "up" | "down" | "click" | "double_click" }
Response: { "ok": true }
Actions map to the glasses touchpad:
up / down — scroll through list items or text
click — select the current item
double_click — triggers a system-level double-click event (typically "back" or "dismiss")
Automation workflow
- Verify the server is running:
GET /api/ping
- Poll for app ready (without clearing console first): look for a known log message.
- Observe: Use
/api/screenshot/glasses (RGBA) or /api/console to check state.
- Act: Send input via
POST /api/input. After each action, wait briefly (~300ms) then confirm state changed.
- Iterate: Use
since_id when polling console to avoid re-reading old entries.
Tips
- The glasses display is 576×288 pixels, monochrome green. Screenshots are RGBA PNGs.
- Input only works when the app has created an active event container. If no container is active, input is silently ignored.
double_click is typically used to go back or dismiss the current view.
- Allow 4+ seconds after launching the simulator before polling — SDK init and
createStartUpPageContainer take time.
Task
$ARGUMENTS