The Agent Skills Directory

[COMMAND_EXECUTION]: The bundled test runner (runner/run.ts) and validation script (runner/check.ts) execute shell commands defined within the project's index.md configuration. These commands are used for stack initialization (backend.start), health checks, and platform-specific test runners (e.g., Playwright or Maestro). Commands are executed using the node:child_process.spawn API with the shell enabled.
[DYNAMIC_EXECUTION]: The runner uses dynamic import() statements to load user-defined TypeScript modules from the project's support/ directory. This mechanism is used to integrate project-specific logic for state management (support/states/*.ts) and persona session handling (support/personas.ts) into the testing lifecycle.
[EXTERNAL_DOWNLOADS]: The skill's reference documentation (references/configuring-ci.md) and contract templates provide examples for setting up GitHub Actions workflows that install the Maestro CLI using a curl | bash command from its official domain (get.maestro.mobile.dev). While this pattern involves remote script execution, it targets a well-known service for its intended functionality.
[INDIRECT_PROMPT_INJECTION]: The skill consumes Markdown test definitions that drive the execution logic and provide parameters for test cases. This creates a surface where malicious instructions embedded in test specifications could influence the agent's behavior or execution flow.
Ingestion points: Markdown files for segments, journeys, and properties located in the project's tests directory (processed by loadSegments, loadJourneys, and loadProperties in runner/model.ts).
Boundary markers: Absent. Test definitions are parsed and their metadata is directly used to construct execution commands.
Capability inventory: The system can execute shell commands, read and write files in the tests root, and dynamically import project-specific code.
Sanitization: The runner includes a quote utility that applies POSIX-style single-quoting to parameter values before they are interpolated into shell command templates, mitigating basic shell injection through parameter values.

building-tests