The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The file scripts/test_runner.py executes system commands using subprocess.run. It determines the command to run based on the contents of a package.json file in the provided project path. An attacker could craft a malicious project where the 'test' script executes harmful commands, which this script would then trigger without further validation.
Dynamic Execution (MEDIUM): The test_runner.py script dynamically constructs command lists (e.g., ['npm', 'test'] or ['npx', 'jest', '--coverage']) based on environment detection. This pattern of building and executing commands from external configuration files is a medium-risk dynamic execution vector as it relies on untrusted metadata to define execution paths.
Indirect Prompt Injection (LOW): The skill is vulnerable to indirect prompt injection via the project files it processes.
Ingestion points: Reads package.json, pyproject.toml, and requirements.txt in scripts/test_runner.py to determine execution logic.
Boundary markers: None; the script lacks delimiters or warnings to ignore malicious instructions embedded in project metadata.
Capability inventory: subprocess.run in scripts/test_runner.py provides shell execution capability via the local environment's test runners.
Sanitization: No sanitization of project-defined scripts or command-line arguments extracted from the target project is performed before execution.

testing-patterns