mobile-automator 0.28.0 → 0.28.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mobile-automator",
-  "version": "0.28.0",
+  "version": "0.28.2",
   "description": "Host-agnostic mauto CLI for AI-driven mobile QA automation — analyzes a mobile project and drives devices through mobile-mcp.",
   "homepage": "https://github.com/sh3lan93/mobile-automator#readme",
   "repository": {
@@ -41,8 +41,7 @@
   "devDependencies": {
     "jest": "^29.7.0",
     "jest-environment-jsdom": "^29.7.0",
-    "js-yaml": "^4.3.2",
-    "toml": "^3.0.0"
+    "js-yaml": "^4.3.2"
   },
   "engines": {
     "node": ">=22.12.0"

package/src/cli.js

@@ -1,6 +1,7 @@
 'use strict';
 
 const fs = require('fs');
+const path = require('path');
 const { Command } = require('commander');
 
 const { version: PKG_VERSION } = require('../package.json');
@@ -190,9 +191,12 @@ async function handleElements({ deviceBridge }) {
   }
 }
 
-async function handleScreenshot({ deviceBridge }, destPath) {
+// Relative paths resolve against the caller's cwd HERE, not in the daemon: the
+// daemon runs from the project root, which is not where a verb run from a
+// subdirectory meant `./shot.png` (#214).
+async function handleScreenshot({ deviceBridge, cwd = process.cwd() }, destPath) {
   try {
-    const savedPath = await deviceBridge.screenshot(destPath);
+    const savedPath = await deviceBridge.screenshot(path.resolve(cwd, destPath));
     return { envelope: ok({ path: savedPath }), exitKind: 'ok' };
   } catch (err) {
     return {
@@ -380,7 +384,10 @@ async function handleLaunch({ deviceBridge }, appId) {
   }
 }
 
-async function handleInstall({ deviceBridge }, appPath) {
+// Resolved against the caller's cwd for the same reason as handleScreenshot
+// (#214). The empty-path guard runs first: path.resolve(cwd, '') is the cwd
+// itself, which would turn a missing argument into a bogus install attempt.
+async function handleInstall({ deviceBridge, cwd = process.cwd() }, appPath) {
   const p = String(appPath == null ? '' : appPath);
   if (!p) {
     return {
@@ -388,9 +395,10 @@ async function handleInstall({ deviceBridge }, appPath) {
       exitKind: 'invalid_input',
     };
   }
+  const abs = path.resolve(cwd, p);
   try {
-    await deviceBridge.installApp(p);
-    return { envelope: ok({ installed: p }), exitKind: 'ok' };
+    await deviceBridge.installApp(abs);
+    return { envelope: ok({ installed: abs }), exitKind: 'ok' };
   } catch (err) {
     return deviceFail(err);
   }
@@ -1807,7 +1815,7 @@ function buildProgram(deps = {}) {
     .command('install <path>')
     .description('Install an app from a local .apk/.app/.ipa path')
     .option('--device <id>', 'target device id')
-    .action(withEnvelope((path, opts) => withBridge(opts.device, (bridge) => handleInstall({ deviceBridge: bridge }, path))));
+    .action(withEnvelope((appPath, opts) => withBridge(opts.device, (bridge) => handleInstall({ deviceBridge: bridge }, appPath))));
 
   program
     .command('uninstall <appId>')

package/src/device/engine-hints.js

@@ -66,6 +66,14 @@ function deviceNotFoundHint(id) {
   );
 }
 
+// B3 (#214): mobile-mcp's validateOutputPath only allows screenshots under the
+// engine's cwd or the temp dir, and its message calls the former "the current
+// directory". The engine runs from the project root (the daemon is spawned
+// there), which is not necessarily the directory the user ran the verb from.
+const OUTSIDE_ALLOWED_DIRS_HINT =
+  'Screenshots must be saved under the project root (the directory containing mobile-automator/) ' +
+  'or the system temp directory. Choose a path inside one of them and retry.';
+
 function engineHint(message, { mobilecliVersion = resolveMobilecliVersion } = {}) {
   try {
     if (typeof message !== 'string' || !message) return null;
@@ -74,6 +82,7 @@ function engineHint(message, { mobilecliVersion = resolveMobilecliVersion } = {}
     }
     const notFound = /Device "([^"]+)" not found/.exec(message);
     if (notFound) return deviceNotFoundHint(notFound[1]);
+    if (/not in the list of allowed directories/.test(message)) return OUTSIDE_ALLOWED_DIRS_HINT;
     return null;
   } catch {
     return null;

package/src/device/mobile-mcp-client.js

@@ -46,6 +46,31 @@ function resolveServerEntry() {
   return path.join(pkgDir, binRel);
 }
 
+// The environment mauto's own engine process is spawned with. Pure: returns a
+// new object and never mutates baseEnv.
+//
+// MOBILEMCP_ALLOW_UNSAFE_URLS defaults to "1" (#212). mobile-mcp rejects every
+// non-http(s) URL in mobile_open_url unless it is set, which blocks the
+// documented `open_url` deep-link capability (myapp://...). That guard exists to
+// stop an LLM whose ONLY reach is MCP from opening arbitrary schemes; an agent
+// driving mauto already has a shell and could open the same URL via adb or
+// simctl, so under mauto it protects nothing. The URL reaches the device as an
+// argv element on the default mobilecli robot (no shell), and the legacy adb
+// robot escapes shell metacharacters. A user-set value wins: the engine only
+// accepts the exact string "1", so MOBILEMCP_ALLOW_UNSAFE_URLS=0 restores the
+// guard. The env is read when the engine spawns, so a running session daemon
+// must be ended (`mauto session end`) before a change takes effect.
+function buildEngineEnv({ device = null, baseEnv = process.env } = {}) {
+  const env = {
+    ...baseEnv,
+    MOBILEMCP_ALLOW_UNSAFE_URLS: baseEnv.MOBILEMCP_ALLOW_UNSAFE_URLS ?? '1',
+  };
+  // mobile-mcp reads the target device from the environment / its own
+  // selection; we pass it through when provided so callers can pin a device.
+  if (device) env.MOBILE_MCP_DEVICE = device;
+  return env;
+}
+
 // The SDK transport, remembering the engine pid the moment the child is spawned
 // (#156). It has to be remembered: on a failed connect the SDK runs
 // `void this.close()`, and StdioClientTransport.close() clears its pid
@@ -84,9 +109,7 @@ async function createCall({
   const transport = new Transport({
     command,
     args,
-    // mobile-mcp reads the target device from the environment / its own
-    // selection; we pass it through when provided so callers can pin a device.
-    env: device ? { ...process.env, MOBILE_MCP_DEVICE: device } : process.env,
+    env: buildEngineEnv({ device }),
   });
 
   const client = new sdk.Client({ name: 'mauto', version: '0.1.0' }, { capabilities: {} });
@@ -255,4 +278,5 @@ module.exports = {
   resolveServerEntry,
   CONNECT_TIMEOUT_MS,
   ENGINE_REAP_MAX_MS,
+  buildEngineEnv,
 };

package/src/device/session-spawn.js

@@ -54,7 +54,12 @@ async function spawnDaemon({
   try {
     // The root in argv is an identity marker only (see paths.daemonArgv): env
     // stays the daemon's source of truth, but `ps` can only see argv.
+    // cwd: the engine inherits the daemon's cwd, and mobile-mcp's
+    // validateOutputPath only allows screenshots under its own process.cwd()
+    // (or the temp dir). Pin it to the project root so the allow-list never
+    // depends on which subdirectory happened to spawn the daemon (#214).
     const child = spawn(process.execPath, [DAEMON_BIN, ...paths.daemonArgv(projectRoot)], {
+      cwd: projectRoot,
       detached: true,
       stdio: log.stdio,
       env,

package/src/guide/content/generate.agnostic.md

@@ -112,7 +112,7 @@ Translate user language to action types and `mauto` verbs:
 | **"allow permission", "grant the permission"** | **`grant_permission`** | **`mauto press grant_permission`** | **Semantic — set optional `permission_name` if specific.** |
 | **"deny permission", "don't allow"** | **`deny_permission`** | **`mauto press deny_permission`** | **Semantic.** |
 | "navigate to [tab]" | `tap` | `mauto elements` to find the tab, then `mauto tap --at <x,y>` | |
-| "open URL", "navigate to URL" | `open_url` | `mauto press` / open the URL | |
+| "open URL", "open deep link", "navigate to URL" | `open_url` | `mauto open-url <url>` | |
 | "wait until visible" | `wait_for_element` | Poll `mauto elements` until present | Set `wait_config.type: "element_visible"` |
 | "wait until gone" | `wait_for_element_gone` | Poll `mauto elements` until absent | |
 | "wait until loaded", "wait for shimmer to stop" | `wait_for_loading_complete` | Poll `mauto screenshot <path>` + visual check | Set `wait_config.indicator` to match project's loading style |

package/src/guide/content/generate.aware.md

@@ -241,7 +241,7 @@ Translate user language to `mauto` verbs and schema actions:
 | "scroll to", "scroll until visible", "scroll down to find" | `scroll_to_element` | `mauto swipe --direction <dir>` (repeated) | Target is the element to scroll to |
 | "go back", "press back" | `press_button` | `mauto press <button>` | Use the back button |
 | "navigate to [tab]" | `tap` | `mauto elements` to find the tab, then `mauto tap --at <x,y>` | |
-| "open URL", "navigate to URL" | `open_url` | `mauto press` / open the URL | |
+| "open URL", "open deep link", "navigate to URL" | `open_url` | `mauto open-url <url>` | |
 | "wait until visible", "wait for [element] to appear" | `wait_for_element` | Poll `mauto elements` until present | Set `wait_config.type: "element_visible"` |
 | "wait until gone", "wait for [element] to disappear" | `wait_for_element_gone` | Poll `mauto elements` until absent | Set `wait_config.type: "element_gone"` |
 | "wait until loaded", "wait for shimmer to stop", "wait for loading to complete" | `wait_for_loading_complete` | Poll `mauto screenshot <path>` + visual check | Set `wait_config.indicator` to match project's loading style (shimmer/spinner/skeleton) |