tether-android-qa
/tether-android-qa
Drive and test a real Android phone or emulator through Tether (adb in the browser) - read the screen, tap by text or id, type, swipe, launch apps, read logcat, run Tether JSON, Maestro or Midscene YAML flows (on one or every connected device), check performance, Perfetto traces and LeakCanary leaks, and record the screen. Use when asked to test, reproduce a bug, or automate something on an Android device. Trigger on "android qa", "adb", "tether", "test on phone", "device farm", "run maestro flow", "logcat", "perf budget", "perfetto trace", "leakcanary", "screen record android".
Assessment and security
- Files in this package
- 3
- Scripts
- 1
- HTML comments
- 0
- URLs in SKILL.md
- 1
No signals detected. Repository changed on 10.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
- 1 scripts in this package. Review them before installing, as Claude may execute them.
Scripts: scripts/pushlab_cli.py
What it does and when to use it
Drive and test a real Android phone or emulator through Tether (adb in the browser) - read the screen, tap by text or id, type, swipe, launch apps, read logcat, run Tether JSON, Maestro or Midscene YAML flows (on one or every connected device), check performance, Perfetto traces and LeakCanary leaks, and record the screen. Use when asked to test, reproduce a bug, or automate something on an Android device. Trigger on "android qa", "adb", "tether", "test on phone", "device farm", "run maestro flow", "logcat", "perf budget", "perfetto trace", "leakcanary", "screen record android".
Triggers from the description
Original author description
Drive and test a real Android phone or emulator through Tether (adb in the browser) - read the screen, tap by text or id, type, swipe, launch apps, read logcat, run Tether JSON, Maestro or Midscene YAML flows (on one or every connected device), check performance, Perfetto traces and LeakCanary leaks, and record the screen. Use when asked to test, reproduce a bug, or automate something on an Android device. Trigger on "android qa", "adb", "tether", "test on phone", "device farm", "run maestro flow", "logcat", "perf budget", "perfetto trace", "leakcanary", "screen record android".
How it works
- Before you start
- Core loop: look, act, verify
- Apps and logs
- Performance
- Repeatable tests (flows)
- Fuzzy goals: AI steps, then deterministic flows
- Evidence for bug reports
- MCP alternative
SKILL.md preview
---
name: tether-android-qa
description: Drive and test a real Android phone or emulator through Tether (adb in the browser) - read the screen, tap by text or id, type, swipe, launch apps, read logcat, run Tether JSON, Maestro or Midscene YAML flows (on one or every connected device), check performance, Perfetto traces and LeakCanary leaks, and record the screen. Use when asked to test, reproduce a bug, or automate something on an Android device. Trigger on "android qa", "adb", "tether", "test on phone", "device farm", "run maestro flow", "logcat", "perf budget", "perfetto trace", "leakcanary", "screen record android".
---
# Tether Android QA
> OpenCode / Sisyphus port of the `tether-android-qa` Claude Code skill from
> `apps/pushlab/plugin/`. Same workflow, same CLI; description extended with
> trigger phrases for OpenCode's skill picker.

Tether exposes a connected Android device (USB, Wi-Fi adb or emulator) over a
local HTTP API. `scripts/pushlab_cli.py` (Python 3 standard library only) wraps
it; every command prints JSON and exits 0 on success, 1 when a check fails and
2 when Tether is unreachable or the request was invalid.
## Before you start
1. Run `python scripts/pushlab_cli.py status`.
- `tether.error` means Tether is not running. Ask the user to start it
(`python apps/pushlab/pushlab.py` starts Tether + Beacon together, or
`python apps/pushlab/tether/server/server.py`) or to set `TETHER_URL` and `TETHER_TOKEN`.
- `tether.devices` must list a device with `state: "device"`. `unauthorized`
means the user must accept the USB-debugging prompt on the phone.
2. With more than one device, pass `--serial <serial>` before the command.
## Core loop: look, act, verify
1. **Look** - `ui` lists on-screen elements (text, resource_id, desc, bounds).
Use `screenshot --out screen.png` and open the image when layout or visuals
matter (colours, overlap, clipped text).
2. **Act** - prefer selectors over coordinates:
- `tap-text "Sign in"`, `tap-id login`, `tap-desc "Navigate up"`
- `type "user@example.com"` types into the focused field (tap it first)
- `key back|home|enter|recents`, `swipe up|down|left|right`
- `tap 0.5 0.8` uses normalised coordinates (0..1) only when nothing is labelled
3. **Verify** - call `ui` (or take a screenshot) after every action. Never assume
a tap worked; check for the expected text or screen.
## Apps and logs
- `apps` lists third-party packages and the foreground activity.
- `launch <package>`, `stop <package>`, `clear <package>` (wipes app data).
- `logcat --package <package> --lines 300` for that app's log; search it for
`FATAL EXCEPTION`, `ANR` and `E/` lines when something misbehaves.
## Performance
- `perf <package>`: memory (PSS), frames (jank %, 90th percentile frame time) and CPU share.
- `perf <package> --startup --runs 3`: cold launch time (median); add `--warm` for a warm start.
- To measure one scenario: `perf <package> --reset`, perform it (scroll, open a screen),
then `perf <package> --max-jank 10 --max-p90-ms 24` — exit 1 means over budget.
- In flows use `perf_startup` / `perf_reset` / `perf_assert` steps (see Tether README).
Report numbers with the device model; emulator numbers only show trends.
- When a budget fails, find out why: `trace start --package <package>`, reproduce,
`trace stop --out slow.pftrace`, and tell the user to open it in https://ui.perfetto.dev.
In flows, `{"action": "perf_trace_start", "keep": "on_fail"}` keeps a trace only when the flow fails.
- Memory leaks: if the app's debug build includes LeakCanary, `leaks --max 0` exits 1 and
names the leaking classes. In flows: `leak_mark`, exercise the screen, go back, then
`{"action": "assert_no_leaks", "wait_ms": 20000}`.
## Repeatable tests (flows)
Write the steps as a Maestro YAML flow (portable to Maestro CLI and Maestro
Cloud) or a Tether JSON flow, then run `flow login.yaml`. Exit code 1 means a
step failed; the JSON report names the step and the reason. Example:
```yaml
appId: com.example.app
---
- launchApp:
clearState: true
- tapOn: "Email"
- inputText: "qa@example.com"
- tapOn:
id: "login"
- assertVisible: "Dashboard"
```
Supported Maestro commands include launchApp (with clearState), tapOn,
doubleTapOn, longPressOn, inputText, eraseText, assertVisible, assertNotVisible,
scroll, scrollUntilVisible, swipe, back, pressKey, hideKeyboard, openLink,
extendedWaitUntil, waitForAnimationToEnd and takeScreenshot; anything else is
skipped and reported in `yaml_warnings`.
Midscene YAML scripts (`tasks:` with `aiAct`, `aiTap`, `aiInput`, `aiAssert`,
`aiWaitFor`, `sleep` ...) run the same way with `flow script.yaml`; their AI steps
need Tether's AI engine configured.
To check several phones or OS versions at once, add `--all-devices`: the flow runs
on every connected device in parallel and the JSON has one report per device
(`passed_devices`, `failed_devices`).
## Fuzzy goals: AI steps, then deterministic flows
When the path through the app is unknown, let an AI step find it first:
- `ai "Book the earliest slot for tomorrow"` (Tether's built-in agent, any
OpenAI-compatible model such as a local Ollama), or add `--engine artemis
--profile pro` to use Google's ARTEMIS agent if its daemon is running.
- `ai "The booking confirmation shows 9:00" --check` judges the screen (PASS/FAIL).
Then follow the ARTEMIS testing mindset: explore with AI, but deliver a
deterministic flow. Re-read the screens with `ui`, write the discovered path as
Maestro YAML with explicit selectors and waits, and run it with `flow` until it
passes repeatedly. AI steps are slow and non-deterministic; keep them for
exploration and visual checks, not for the regression suite.
## Evidence for bug reports
- `record start` then reproduce, then `record stop --out bug.mp4` (MP4 from
`adb screenrecord`, max 180 s, no audio).
- Attach the screenshot, the failing flow report and the relevant logcat lines.
## MCP alternative
Tether also ships an MCP server (`apps/pushlab/tether/server/mcp_server.py`) with
mobile-mcp compatible tools (`mobile_list_elements_on_screen`,
`mobile_click_on_screen_at_coordinates`, ...). If those tools are available in
this session, you may use them instead of the CLI; the workflow is the same.
Files in this package (3)
- SKILL.md 6 KB
- illustration.svg 9 KB
- scripts/pushlab_cli.py 28 KB
Versions
The database has no other skills named tether-android-qa.
Database snapshot: 13.11.2026 · package version cf5bbc9652b6 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100tethertrace
Test Android apps on the user's real phone over Wi-Fi with tethertrace (no adb, no Android SDK). Use when the user wants to install an APK on their phone, check an Android build for crashes, explore an app automatically, write or run a YAML UI test, see what is on the phone's… - 100mobiai-android-device
Use when interacting with an Android device or emulator — run adb commands, automate UI, capture screenshots, read logcat, manage emulators. - 100push-qa-e2e
End-to-end push notification QA on a real Android device - subscribe the phone to Beacon through Tether, send a push, assert it appears in the notification shade, tap it and verify the app or page it opens, with screen recording as evidence. Use when asked to verify that push… - 85android-emulator-qa
Use when validating Android feature flows in an emulator with adb-driven launch, input, UI-tree inspection, screenshots, and logcat capture. - 90test-android-app
Test an Android app on a real cloud Android device (ApkRunner) - build the APK, install it, walk through the flows the user names like a manual tester, collect crashes and logcat, and report what works and what breaks. Use when asked to run, try, smoke-test or QA an Android app… - 90phone-use
Use OpenPocket MCP tools to inspect or control an Android emulator, physical Android phone, Android TV, or ADB-backed app. Trigger for phone use, Android automation, mobile UI testing, emulator control, physical-device control, app navigation, screenshots, tapping, typing, and…