The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided project context to generate test plans, creating a surface for indirect prompt injection.
Ingestion points: The skill accepts "app scope, platform details, release notes, and user flows" as input for designing test cases in prompts/mobile-testing.md.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present in the prompts or scripts.
Capability inventory: The agent can invoke various Python scripts in the scripts/ directory to convert test artifacts into different formats (Word, Excel, JSON, etc.) and write them to the local filesystem.
Sanitization: No sanitization or validation of the untrusted project data is performed before it is used to generate testing scenarios.
[COMMAND_EXECUTION]: The skill relies on multiple Python scripts that utilize the subprocess module to execute other internal conversion and parsing tools.
Evidence: scripts/batch_convert_templates.py, scripts/convert_to_csv.py, and several other formatting wrappers execute convert_formats.py or parse_formats.py using subprocess.call() with arguments passed from the command line.
[PRIVILEGE_ESCALATION]: The documentation includes setup instructions that suggest lowering the system's security posture by bypassing local script execution policies.
Evidence: README.md instructs the user to run PowerShell with the -ExecutionPolicy Bypass flag to execute the skill's installation script.

mobile-testing