android-emu-agent 0.1.13

CLI + daemon for LLM-driven Android UI control — ships with ready-to-use coding agent skills

Android Emu Agent

CLI + daemon for LLM-driven Android UI control — ships with ready-to-use coding agent skills.

Documentation | CLI reference | Source

Overview

Android Emu Agent automates Android apps using a fast observe-act-verify loop:

1. Observe: capture a compact actionable UI snapshot
2. Act: issue commands using ephemeral element refs like ^a1
3. Verify: re-snapshot when needed

The CLI is a thin client. A long-running daemon handles all device I/O. All commands support --json for machine-readable output, making the tool ideal for agent consumption.

Highlights:

• Daemon-first architecture — persistent process handles device I/O over Unix socket
• Framework-aware snapshots — actionable output works across classic XML Views and modern UI frameworks such as Compose and Litho
• Ref healing — deterministic ^a1-style handles with selector-chain rebinding when a newer snapshot still contains the same target
• Richer selectors — exact, contains, regex, fallback, state-filter, resource ID, content-desc, class, and coordinate selectors with capability introspection
• Diagnostics — JSON responses and headers include diagnostic_id for request-level tracing
• Trace archives — record daemon exchanges into replayable .aea-trace.zip evidence bundles
• Task harness — run JSON task specs or human-editable .aea scripts with step-level and final verifiers
• Expectations — assertion-style commands for UI state, activity, and foreground app checks
• Visual grounding — optional screenshot-to-ref metadata for human or vision-model evidence
• System surfaces — open notifications/Quick Settings and list, grant, or revoke app permissions through validated shell-backed commands
• Debugger fusion — combine current app state, latest UI refs, debugger status, events, logpoint hits, and stack in one bounded observation
• Performance profile — aggregate process, memory, rendering, background, exit, and event signals into one app health snapshot
• Native performance artifacts — capture bounded Perfetto traces, simpleperf CPU profiles, and screen recordings as local evidence
• Agent skills included — structured reference docs, workflow templates, and safety guardrails
• Machine-readable output — every command supports --json for agent pipelines

Inspiration

Inspired by agent-browser, a fast headless browser automation CLI for AI agents.

Requirements

• Python 3.11+
• uv package manager
• Android SDK with adb, emulator, and ideally avdmanager on PATH
• Android emulator or rooted device (primary target)
• JDK 17+ (only for debugger commands; set JAVA_HOME or have java on PATH)

Install

# Clone the repository
git clone https://github.com/alehkot/android-emu-agent.git
cd android-emu-agent

# Install dependencies
uv sync

Inside this repo, prefer uv run android-emu-agent <command>.

Android SDK CLI Prerequisites

For emulator lifecycle commands, the daemon expects Android SDK command-line tools to be available. It checks PATH first and then common SDK roots derived from ANDROID_SDK_ROOT / ANDROID_HOME.

Recommended PATH entries:

• platform-tools for adb
• emulator for emulator
• cmdline-tools/latest/bin for avdmanager and sdkmanager

Example macOS setup:

export ANDROID_SDK_ROOT="$HOME/Library/Android/sdk"
export PATH="$ANDROID_SDK_ROOT/platform-tools:$ANDROID_SDK_ROOT/emulator:$ANDROID_SDK_ROOT/cmdline-tools/latest/bin:$PATH"

adb version
emulator -list-avds
avdmanager list avd

If you only plan to control an already-running device, adb is enough. If you want android-emu-agent to boot or manage emulator instances, make emulator available too.

Quick Start (2 minutes)

# 0. Optional: boot an AVD from the CLI
uv run android-emu-agent emulator list-avds
uv run android-emu-agent emulator start Pixel_8_API_34 --wait-boot

# 1. Start the daemon (optional, CLI will auto-start by default)
uv run android-emu-agent daemon start

# 2. List connected devices
uv run android-emu-agent device list

# 3. Start a session (prints the session id)
uv run android-emu-agent session start --device emulator-5554

# 4. Take a compact snapshot and read refs
uv run android-emu-agent ui snapshot s-abc123 --format text

# 5. Tap an element using a ref
uv run android-emu-agent action tap s-abc123 ^a1

# 6. Stop the session
uv run android-emu-agent session stop s-abc123

Most commands accept --json for machine-readable output.

Example --json output for session start:

{
  "status": "done",
  "session_id": "s-abc123",
  "device_serial": "emulator-5554",
  "generation": 0
}

Agent Skills

This repo ships a first-class android-emu-agent skill in skills/android-emu-agent/ for coding agents that support skills (Claude Code, Codex, and similar environments).

The skill provides:

• Structured reference docs — command reference, troubleshooting, error codes
• Workflow templates — screen inspection, login flows, navigation patterns, form filling
• Recovery protocols — handling stale refs, dialog blockers, idle waits
• Reliability guidance — crash/ANR triage, performance artifacts, debugger handoff
• Safety guardrails — read-only inquiry handling, root checks, emulator-only safeguards

Use it when an agent needs to inspect an Android screen, interact with an app, run emulator/device UI automation, diagnose missing or stale elements, collect screenshots/logs/traces, investigate crashes or performance regressions, push/pull device files, or use the debugger bridge.

Install via dev script (recommended)

./scripts/dev.sh skills          # Symlink to all supported agents
./scripts/dev.sh skills claude   # Claude Code only
./scripts/dev.sh skills codex    # Codex only
./scripts/dev.sh skills vscode   # VS Code project-local .agents/skills
./scripts/dev.sh skills-validate # Validate bundled skill structure and references

Manual install

VS Code / project-local Agent Skills:

mkdir -p .agents/skills
ln -sfn "$(pwd)/skills/android-emu-agent" .agents/skills/android-emu-agent

Claude Code:

mkdir -p ~/.claude/skills
ln -sfn "$(pwd)/skills/android-emu-agent" ~/.claude/skills/android-emu-agent

Codex:

export CODEX_HOME="$HOME/.codex"
mkdir -p "$CODEX_HOME/skills"
ln -sfn "$(pwd)/skills/android-emu-agent" "$CODEX_HOME/skills/android-emu-agent"

If symlinks are not an option, copy the directory instead.

Using the Skill

After installation, the agent should be primed to start a session with your connected device before you ask for specific actions. Begin with a direct initialization request, for example:

I want to interact with my connected Android emulator.

Once the agent has initialized the session, you can proceed with normal requests (snapshots, tap/type actions, app launches, waits, etc.).

Prerequisites you may need (if the agent reports it cannot connect):

1. If no emulator is running yet, list and boot one: uv run android-emu-agent emulator list-avds uv run android-emu-agent emulator start <avd_name> --wait-boot
2. Start the daemon: uv run android-emu-agent daemon start
3. Confirm a device is visible: uv run android-emu-agent device list
4. If needed, start a session explicitly: uv run android-emu-agent session start --device emulator-5554

Core Concepts

Sessions

• Sessions tie actions to a specific device. Most commands accept a session id.
• session start returns a session id. session stop releases it.

Snapshots and refs

• ui snapshot returns actionable ^refs that are generation-scoped and optimized for agent use.
• Compact snapshots are designed to work well with classic XML layouts and modern frameworks such as Compose and Litho.
• If a ref is stale, the daemon may heal it against the latest snapshot and return a warning. You should still take a fresh snapshot before the next action.

Selectors

• action tap accepts an ^ref or a selector.
• long-tap, set-text, and clear require an ^ref.

Selector examples:

^a1
text:"Sign in"
id:com.example:id/login_btn
desc:"Open navigation"
coords:120,450

Snapshot Format

Compact snapshot output includes context and actionable UI elements only:

{
  "schema_version": 1,
  "session_id": "s-abc123",
  "generation": 42,
  "context": {
    "package": "com.example.app",
    "activity": ".MainActivity",
    "orientation": "PORTRAIT",
    "ime_visible": false
  },
  "elements": [
    {
      "ref": "^a1",
      "role": "button",
      "label": "Sign in",
      "resource_id": "com.example:id/login_btn",
      "bounds": [100, 200, 300, 250],
      "state": { "clickable": true, "enabled": true }
    }
  ]
}

Compact mode may proxy descendant text onto an unlabeled clickable container when that makes the target easier to identify. Use --full for the unpruned hierarchy view, or --raw for the raw XML tree.

Daemon and State

• Socket: /tmp/android-emu-agent.sock
• Logs: ~/.android-emu-agent/daemon.log
• PID file: ~/.android-emu-agent/daemon.pid
• Request diagnostics: ~/.android-emu-agent/diagnostics/requests.ndjson

The CLI auto-starts the daemon on first request. Use these to debug:

uv run android-emu-agent daemon status --json
uv run android-emu-agent daemon stop
uv run android-emu-agent daemon start

Common Workflows

Login flow

uv run android-emu-agent app launch s-abc123 com.example.app
uv run android-emu-agent wait idle s-abc123 --timeout-ms 5000
uv run android-emu-agent ui snapshot s-abc123 --format text
uv run android-emu-agent action set-text s-abc123 ^a3 "[e-mail]"
uv run android-emu-agent action set-text s-abc123 ^a4 "hunter2"
uv run android-emu-agent action tap s-abc123 ^a5

When elements are missing

uv run android-emu-agent ui snapshot s-abc123 --full
uv run android-emu-agent wait idle s-abc123 --timeout-ms 3000
uv run android-emu-agent ui snapshot s-abc123

Inspect capabilities and use richer selectors

uv run android-emu-agent device capabilities --session s-abc123 --json
uv run android-emu-agent action tap s-abc123 'text-contains:"Continue"'
uv run android-emu-agent wait exists s-abc123 --id-matches '.*checkout.*'
uv run android-emu-agent expect exists s-abc123 --class android.widget.Button --text-contains Pay

Assert expected state

uv run android-emu-agent expect text s-abc123 "Welcome" --timeout-ms 5000
uv run android-emu-agent expect exists s-abc123 --text "Checkout" --timeout-ms 5000
uv run android-emu-agent expect current-app s-abc123 --package com.example.app --activity CheckoutActivity

Visual debug

uv run android-emu-agent ui screenshot --device emulator-5554 --pull --output ./screen.png
uv run android-emu-agent ui ground s-abc123 --ref ^a1 --pull --output ./grounding.json
uv run android-emu-agent artifact bundle s-abc123
uv run android-emu-agent artifact logs --session s-abc123 --app com.example.app --type errors --since "10m ago"

System surfaces and permissions

uv run android-emu-agent system notifications open --session s-abc123
uv run android-emu-agent system quick-settings open --session s-abc123
uv run android-emu-agent system permissions list com.example.app --session s-abc123
uv run android-emu-agent system permissions grant com.example.app android.permission.POST_NOTIFICATIONS --session s-abc123
uv run android-emu-agent system permissions revoke com.example.app android.permission.POST_NOTIFICATIONS --session s-abc123

Trace an agent run

uv run android-emu-agent trace start s-abc123 --label checkout-repro
uv run android-emu-agent action tap s-abc123 ^a1
uv run android-emu-agent ui snapshot s-abc123
uv run android-emu-agent trace stop s-abc123 --output ./artifacts/checkout-repro.aea-trace.zip
uv run android-emu-agent trace replay ./artifacts/checkout-repro.aea-trace.zip --until-failure
uv run android-emu-agent trace export ./artifacts/checkout-repro.aea-trace.zip --output ./artifacts/checkout-repro.md

Run a task harness spec

{
  "name": "checkout smoke",
  "session_id": "s-abc123",
  "steps": [
    {
      "name": "open checkout",
      "action": "tap",
      "ref": "text:\"Checkout\"",
      "verify": [{ "type": "exists", "text": "Payment" }]
    }
  ],
  "verifiers": [{ "type": "activity", "activity": "CheckoutActivity" }]
}

uv run android-emu-agent task validate ./checkout-task.json
uv run android-emu-agent task run ./checkout-task.json --json

Or use a human-editable .aea task script:

name "checkout smoke"
session s-abc123
tap text:"Checkout" || id:com.example:id/checkout
verify exists text:"Payment"
expect activity CheckoutActivity

uv run android-emu-agent task validate ./checkout-smoke.aea
uv run android-emu-agent task run ./checkout-smoke.aea --json

App debug helpers

uv run android-emu-agent app current --session s-abc123
uv run android-emu-agent app task-stack --session s-abc123
uv run android-emu-agent app resolve-intent --session s-abc123 --action android.intent.action.VIEW --data "https://example.com/deep"
uv run android-emu-agent reliability profile com.example.app --session s-abc123 --json
uv run android-emu-agent reliability process com.example.app --device emulator-5554
uv run android-emu-agent reliability meminfo com.example.app --device emulator-5554
uv run android-emu-agent reliability gfxinfo com.example.app --device emulator-5554
uv run android-emu-agent reliability perfetto --session s-abc123 --duration 10
uv run android-emu-agent reliability simpleperf com.example.app --session s-abc123 --duration 10
uv run android-emu-agent reliability screenrecord --session s-abc123 --duration 10

Debugger (JDI Bridge)

Attach a JDWP debugger to a running Android app's JVM. This requires JDK 17+ and a debuggable app (built with android:debuggable=true or running on a userdebug/eng device).

# Verify bridge startup and JSON-RPC health
uv run android-emu-agent debug ping s-abc123

# Launch the app in wait-for-debugger mode
uv run android-emu-agent app launch s-abc123 com.example.app --wait-debugger

# Attach the debugger (finds PID, sets up ADB forward, connects via JDI)
uv run android-emu-agent debug attach --session s-abc123 --package com.example.app

# Keep the VM suspended after attach (useful with app launch --wait-debugger)
uv run android-emu-agent debug attach --session s-abc123 --package com.example.app --keep-suspended

# If multiple debuggable processes exist, pick one explicitly
uv run android-emu-agent debug attach --session s-abc123 --package com.example.app --process com.example.app:remote

# Check debug session status (VM name, version, thread count)
uv run android-emu-agent debug status --session s-abc123

# Fused app/UI/debug context for agent planning
uv run android-emu-agent debug observe --session s-abc123 --json

# Set/list/remove breakpoints
uv run android-emu-agent debug break set com.example.app.MainActivity 42 --session s-abc123
uv run android-emu-agent debug break list --session s-abc123
uv run android-emu-agent debug break remove 1 --session s-abc123

# List threads (default skips daemon threads; --all includes them)
uv run android-emu-agent debug threads --session s-abc123
uv run android-emu-agent debug threads --session s-abc123 --all

# Drain debugger event queue (breakpoint hits/resolutions, disconnect events)
uv run android-emu-agent debug events --session s-abc123

# Stack and variable inspection on suspended threads
uv run android-emu-agent debug stack --session s-abc123 --thread main --max-frames 10
uv run android-emu-agent debug inspect savedInstanceState --session s-abc123 --thread main --frame 0 --depth 1
uv run android-emu-agent debug eval savedInstanceState.toString() --session s-abc123 --thread main --frame 0

# Optional: load/clear ProGuard-R8 mapping for deobfuscated names in stack/inspect output
uv run android-emu-agent debug mapping load ./mapping.txt --session s-abc123
uv run android-emu-agent debug mapping clear --session s-abc123

# Step execution (observe-act-verify): returns new location + locals atomically
uv run android-emu-agent debug step-over --session s-abc123 --thread main
uv run android-emu-agent debug step-into --session s-abc123 --thread main
uv run android-emu-agent debug step-out --session s-abc123 --thread main

# Resume one thread or all threads
uv run android-emu-agent debug resume --session s-abc123 --thread main
uv run android-emu-agent debug resume --session s-abc123

# Detach when done (cleans up ADB forward and bridge process)
uv run android-emu-agent debug detach --session s-abc123

Step commands default to a 10s timeout (--timeout-seconds) and return actionable timeout payloads when a step cannot complete. If the main thread has been suspended for too long, responses include an ANR warning so you can resume before Android's ANR threshold. When a mapping file is loaded, stack class/method names and inspect field names are deobfuscated. Mapping state is per attached bridge instance and clears on detach.

debug observe is non-destructive by default: it peeks queued debugger events, reads buffered logpoint hits, summarizes the latest snapshot refs, and only drains events when --drain-events is set.

Example debugger workflows

Launch suspended, attach, set breakpoints, then resume

# Start app in wait-for-debugger mode (process is paused before app code runs)
uv run android-emu-agent app launch s-abc123 com.example.app --wait-debugger

# Attach and keep VM suspended (do not auto-resume on attach)
uv run android-emu-agent debug attach --session s-abc123 --package com.example.app --keep-suspended

# Set breakpoints while execution is still paused
uv run android-emu-agent debug break set com.example.app.MainActivity 42 --session s-abc123
uv run android-emu-agent debug break set com.example.checkout.CartViewModel 118 --session s-abc123 --condition "cart.total > 10000"

# Optional: verify configured breakpoints before continuing
uv run android-emu-agent debug break list --session s-abc123

# Resume only after breakpoints are in place
uv run android-emu-agent debug resume --session s-abc123

Conditional breakpoint for state-specific bugs

# Attach
uv run android-emu-agent app launch s-abc123 com.example.app --wait-debugger
uv run android-emu-agent debug attach --session s-abc123 --package com.example.app

# Stop only when cart total crosses threshold
uv run android-emu-agent debug break set com.example.checkout.CartViewModel 118 --session s-abc123 --condition "cart.total > 10000"
uv run android-emu-agent debug resume --session s-abc123

# Wait for hit, then inspect state
uv run android-emu-agent debug events --session s-abc123
uv run android-emu-agent debug stack --session s-abc123 --thread main --max-frames 15
uv run android-emu-agent debug inspect cart.total --session s-abc123 --thread main --frame 0
uv run android-emu-agent debug resume --session s-abc123

Condition syntax supports value paths, literals (null, booleans, numbers, strings), boolean operators (!, &&, ||), comparisons (==, !=, >, >=, <, <=), and parentheses. Malformed condition syntax fails fast during debug break set. Runtime evaluation failures emit breakpoint_condition_error and auto-resume.

Non-suspending logpoint trace

# Attach and set a logpoint (does not suspend thread)
uv run android-emu-agent debug attach --session s-abc123 --package com.example.app
uv run android-emu-agent debug break set com.example.login.LoginViewModel 87 --session s-abc123 --log-message "attempt={hitCount} user={username} locked={isLocked}" --capture-stack --stack-max-frames 8

# Keep app running; hits are buffered with timestamps (and optional stack)
uv run android-emu-agent debug resume --session s-abc123
uv run android-emu-agent debug break hits --session s-abc123 --limit 100
uv run android-emu-agent debug break hits --session s-abc123 --breakpoint-id 1

# Clean up
uv run android-emu-agent debug break list --session s-abc123
uv run android-emu-agent debug break remove 1 --session s-abc123

Buffered logpoint example use cases

# 1) High-frequency callback tracing without pausing app execution
uv run android-emu-agent debug break set com.example.sync.SyncWorker 214 --session s-abc123 --log-message "sync run={hitCount} state={state}" --capture-stack --stack-max-frames 6
uv run android-emu-agent debug resume --session s-abc123
uv run android-emu-agent debug break hits --session s-abc123 --breakpoint-id 1 --limit 50

# 2) Investigate only hits after a known incident timestamp (epoch ms)
uv run android-emu-agent debug break hits --session s-abc123 --since-timestamp-ms 1735689600000 --limit 200

# 3) Compare two non-suspending logpoints independently
uv run android-emu-agent debug break set com.example.login.LoginViewModel 87 --session s-abc123 --log-message "attempt={hitCount} user={username}"
uv run android-emu-agent debug break set com.example.session.SessionStore 55 --session s-abc123 --log-message "refresh={hitCount} token={tokenState}"
uv run android-emu-agent debug break hits --session s-abc123 --breakpoint-id 1 --limit 100
uv run android-emu-agent debug break hits --session s-abc123 --breakpoint-id 2 --limit 100

Crash triage with exception breakpoints

# Attach and break on uncaught exceptions
uv run android-emu-agent app launch s-abc123 com.example.app --wait-debugger
uv run android-emu-agent debug attach --session s-abc123 --package com.example.app
uv run android-emu-agent debug break-exception set --session s-abc123 --class '*' --no-caught --uncaught

# Reproduce crash path, then inspect exception event
uv run android-emu-agent debug resume --session s-abc123
uv run android-emu-agent debug events --session s-abc123
uv run android-emu-agent debug stack --session s-abc123 --thread main --max-frames 20

# Optional: narrow to one exception class after first capture
uv run android-emu-agent debug break-exception remove 1 --session s-abc123
uv run android-emu-agent debug break-exception set --session s-abc123 --class java.lang.NullPointerException --caught --uncaught

Under the hood, the daemon spawns a JDI Bridge sidecar (a Kotlin/JVM subprocess in jdi-bridge/) that speaks JSON-RPC over stdin/stdout. The bridge connects to the target app via JDWP and monitors for VM disconnect events. Build the bridge JAR with:

cd jdi-bridge && ./gradlew shadowJar

Emulator lifecycle and snapshots

uv run android-emu-agent emulator list-avds
uv run android-emu-agent emulator start Pixel_8_API_34 --snapshot clean --no-snapshot-save
uv run android-emu-agent emulator stop emulator-5554
uv run android-emu-agent emulator snapshot save emulator-5554 clean
uv run android-emu-agent emulator snapshot restore emulator-5554 clean

These commands require an emulator serial (emulator-5554). If you pass a non-emulator serial, you will see ERR_NOT_EMULATOR. Snapshot restore restarts the emulator by default so the loaded snapshot becomes the active runtime state again. Use --no-restart if you explicitly want the old live-load behavior. Create AVD definitions ahead of time with Android Studio or avdmanager.

Real Devices (Non-Root)

The project targets emulators and rooted devices, but many commands do not enforce root and can work on real devices as long as adb is connected and uiautomator2 can attach (ATX server on port 7912). In practice, these are usually safe on non-root devices:

• UI snapshots, screenshots, and optional visual grounding
• Actions (tap, set-text, swipe, scroll, back/home/recents)
• Wait commands
• Expect commands
• Device capabilities
• System surfaces (notifications, quick-settings) and runtime permission list/grant/revoke
• App list/install/uninstall/launch/intent/force-stop/reset/deeplink
• Reliability profile, native perf artifacts, events, process, meminfo, gfxinfo, and background diagnostics
• File push and pull to shared storage

Emulator-only commands are emulator snapshot save|restore. Root-required commands are listed below.

Root-Required Operations

The following require a rooted device or emulator with root access:

• reliability oom-adj
• reliability pull anr
• reliability pull tombstones
• reliability pull dropbox
• file find
• file list
• file app push
• file app pull

If root is missing, you will see ERR_PERMISSION.

Artifacts

Artifacts are written to ~/.android-emu-agent/artifacts by default.

• Reliability outputs: ~/.android-emu-agent/artifacts/reliability
• File transfers: ~/.android-emu-agent/artifacts/files
• Visual grounding: ~/.android-emu-agent/artifacts/visual
• Trace archives: ~/.android-emu-agent/traces
• Local repo artifact scratch space: ./artifacts/ (ignored by git)

Troubleshooting

Quick checks

uv run android-emu-agent device list
adb devices
uv run android-emu-agent daemon status --json

Common errors

Error Code	Meaning	Fix
ERR_STALE_REF	Ref from an old snapshot	Re-snapshot; if auto-healed, use warning as cue
ERR_NOT_FOUND	Element not found	Verify screen, use --full or a selector
ERR_BLOCKED_INPUT	Dialog/IME blocking	wait idle or back
ERR_TIMEOUT	Wait condition not met	Increase --timeout-ms or check condition
ERR_DEVICE_OFFLINE	Device disconnected	Reconnect and re-run device list
ERR_SESSION_EXPIRED	Session is gone	Start a new session
ERR_PERMISSION	Root required	Use a rooted device/emulator
ERR_ADB_NOT_FOUND	adb not on PATH	Install Android SDK and ensure adb is on PATH
ERR_SDK_TOOL_NOT_FOUND	SDK CLI tool missing	Add emulator / avdmanager to PATH
ERR_ADB_COMMAND	ADB command failed	Check device connectivity and retry
ERR_ALREADY_ATTACHED	Debug session exists	Detach first with debug detach
ERR_DEBUG_NOT_ATTACHED	No debug session	Attach first with debug attach
ERR_JDK_NOT_FOUND	Java not found	Install JDK 17+ or set JAVA_HOME
ERR_VM_DISCONNECTED	Target VM exited	Re-launch the app and re-attach
ERR_TRACE_ACTIVE	Trace already active	Stop the active trace before starting another
ERR_TRACE_NOT_ACTIVE	No active trace	Start a trace before stopping it
ERR_TRACE_INVALID	Bad trace archive	Use a .aea-trace.zip produced by trace stop
ERR_TASK_INVALID	Bad task spec	Fix JSON and run task validate
ERR_TASK_UNSUPPORTED_STEP	Unsupported task op	Use supported action/wait/app/ui operations
ERR_EXPECTATION_FAILED	Assertion failed	Check state, selector, or timeout
ERR_EXPECTATION_REQUIRED	Missing expected state	Provide --package or --activity
ERR_SNAPSHOT_REQUIRED	Missing UI snapshot	Run ui snapshot before visual grounding
ERR_VISUAL_REF_NOT_FOUND	Ref missing in snapshot	Use refs from latest snapshot generation
ERR_VISUAL_SNAPSHOT_INVALID	Bad snapshot metadata	Re-run ui snapshot before grounding
ERR_INVALID_PERMISSION	Bad permission name	Use a fully-qualified Android permission name
ERR_SYSTEM_COMMAND_FAILED	System shell failed	Check device state and Android service support

For deeper guidance, see skills/android-emu-agent/references/troubleshooting.md.

Every JSON response also includes a diagnostic_id, and the same ID is returned as the x-diagnostic-id header for correlation with daemon request logs.

CLI Reference

Full auto-generated CLI reference is available at docs/reference.md.

To regenerate or browse the docs locally:

./scripts/dev.sh docs-gen      # Regenerate docs/reference.md from CLI
./scripts/dev.sh docs          # Build to site/
./scripts/dev.sh docs-serve    # Serve at http://127.0.0.1:8000

A concise command list also lives at skills/android-emu-agent/references/command-reference.md.

If you prefer an interactive guide:

uv run android-emu-agent --help
uv run android-emu-agent <group> --help

Architecture

┌─────────────────┐     Unix Socket      ┌──────────────────────────────────┐
│                 │  ◄────────────────►  │            Daemon                │
│   CLI Client    │                      │  ┌────────────────────────────┐  │
│                 │                      │  │     Device Manager         │  │
└─────────────────┘                      │  │  (adbutils, uiautomator2)  │  │
                                         │  └────────────────────────────┘  │
                                         │  ┌────────────────────────────┐  │
                                         │  │    Session Manager         │  │
                                         │  │  (refs, state, SQLite)     │  │
                                         │  └────────────────────────────┘  │
                                         │  ┌────────────────────────────┐  │
                                         │  │    UI Snapshotter          │  │
                                         │  │  (lxml, filtering)         │  │
                                         │  └────────────────────────────┘  │
                                         │  ┌────────────────────────────┐  │
                                         │  │    Debug Manager           │  │
                                         │  │  (JDI Bridge, JDWP)       │──┼──► JDI Bridge
                                         │  └────────────────────────────┘  │    (Kotlin subprocess)
                                         └──────────────────────────────────┘
                                                         │
                                                         ▼
                                         ┌──────────────────────────────────┐
                                         │     Android Device/Emulator      │
                                         │  (ATX Server on port 7912)       │
                                         └──────────────────────────────────┘

Development Scripts

The ./scripts/dev.sh helper centralizes common development tasks. Make it executable with chmod +x scripts/dev.sh if needed.

Command	Description
./scripts/dev.sh setup	Install dependencies
./scripts/dev.sh check	Run all checks (lint + typecheck + unit tests + docs)
./scripts/dev.sh test	Run all tests
./scripts/dev.sh test-unit	Run unit tests only
./scripts/dev.sh test-integration	Run integration tests (requires emulator)
./scripts/dev.sh build-bridge	Build the JDI Bridge fat JAR (jdi-bridge/)
./scripts/dev.sh test-bridge	Run JDI Bridge Kotlin tests
./scripts/dev.sh lint	Run linter
./scripts/dev.sh format	Format code
./scripts/dev.sh format-md	Format Markdown
./scripts/dev.sh lint-md	Lint Markdown
./scripts/dev.sh md	Format + lint Markdown
./scripts/dev.sh hooks	Install git hooks
./scripts/dev.sh typecheck	Run type checkers (mypy + pyright)
./scripts/dev.sh daemon	Start the daemon server
./scripts/dev.sh bump-version	Interactively bump version, refresh lock, optionally tag
./scripts/dev.sh release	Build artifacts, publish GitHub release and PyPI
./scripts/dev.sh verify-release	Verify release bridge JAR/checksum assets exist
./scripts/dev.sh publish-pypi	Rebuild checked Python dist and upload with Twine
./scripts/dev.sh docs	Build documentation (mkdocs)
./scripts/dev.sh docs-serve	Serve documentation locally
./scripts/dev.sh docs-gen	Regenerate CLI reference from Typer app
./scripts/dev.sh skills [target]	Symlink skills into agent directories
./scripts/dev.sh skills-validate	Validate bundled skill structure and references
./scripts/dev.sh skills-codex	Symlink skills into Codex agent directory
./scripts/dev.sh skills-vscode	Symlink skills into VS Code .agents/skills
./scripts/dev.sh skills-claude	Symlink skills into Claude agent directory

Raw uv run commands

# Install with dev dependencies
uv sync --all-extras

# Run tests
uv run pytest tests/unit -v

# Run linter
uv run ruff check .

# Format code
uv run ruff format .

# Type check
uv run mypy src/

License

MIT. See LICENSE.