The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install the Maestro CLI by downloading a script from a remote URL and piping it directly into bash.
Evidence in SKILL.md: curl -fsSL "https://get.maestro.mobile.dev" | bash used for tool installation.
Evidence in references/commands/cli-and-artifacts.md: Repeated installation instructions using the same piped-to-bash pattern.
[COMMAND_EXECUTION]: The skill provides extensive templates for orchestrating workspace transfers and executing shell commands on remote macOS hosts using SSH, which involves complex shell interpolation.
Evidence in references/guides/ios-over-ssh.md: Contains a multi-stage shell script recipe that uses Python-based shell quoting (shlex.quote) to build and execute remote commands, including workspace cleanup and driver management.
[DYNAMIC_EXECUTION]: Maestro uses an embedded GraalJS runtime to execute JavaScript for dynamic data generation and pre-flight HTTP requests during test execution.
Evidence in references/guides/graaljs.md: Detailed documentation for evalScript and runScript commands which allow execution of local JavaScript files and inline expressions.
[EXTERNAL_DOWNLOADS]: In addition to the CLI installation, the skill facilitates downloading external binaries and artifacts.
Evidence: Downloads internal dependencies like applesimutils and retrieves artifacts from remote simulators via SSH.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a broad attack surface for indirect injection by ingesting view hierarchies from potentially untrusted mobile applications.
Ingestion points: maestro hierarchy and MCP inspect_screen tool read attributes (text, labels, IDs) from the application under test.
Capability inventory: The skill has the ability to execute shell scripts, write files (artifacts), and perform network requests (GraalJS http client).
Boundary markers: No specific boundary markers or instructions to ignore embedded content are provided when the agent processes the retrieved UI hierarchy.
Sanitization: While the skill uses shlex.quote for its own script generation, there is no explicit sanitization of data extracted from the mobile screen before it is processed by the agent.

test-by-maestro