The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the Maestro CLI executable from its official GitHub repository. This process is handled securely in templates/scripts/e2e/install-maestro.sh using a hardcoded version (2.10.0) and performing mandatory SHA-256 checksum verification (shasum -a 256 -c) before extraction. Per the analysis rules, downloads from well-known platforms like GitHub for established tools are considered safe.
[COMMAND_EXECUTION]: The runner and setup scripts execute various system utilities including adb, xcrun simctl, pgrep, lsof, unzip, and plutil to manage mobile devices and inspect application state. All commands are constructed using internal bindings that enforce strict device targeting (by UDID or adb serial) and specific driver ports, which prevents accidental or malicious interference in shared multi-user environments.
[DATA_EXFILTRATION]: No evidence of data exfiltration was found. The skill's network sampler (templates/scripts/e2e/sample-sockets.ts) monitors the app's own process sockets for the sole purpose of enforcing the repository's network policy. This data is written to a local report file and is not transmitted to external servers.
[SAFE]: The skill explicitly disables all telemetry and analytics for the Maestro toolchain by setting environment variables such as MAESTRO_CLI_NO_ANALYTICS and MAESTRO_DISABLE_UPDATE_CHECK to true in the maestro-run.ts and install-maestro.sh templates. This ensures that the testing process remains private and machine-local.
[INDIRECT_PROMPT_INJECTION]: The skill processes structured configuration (JSON/YAML) and test reports (JUnit XML). It uses a custom YAML parser (maestro-yaml.mjs) and the standard valibot library for schema validation, which limits the attack surface for indirect prompt injection via malformed external data.

e2e-maestro