The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of functional specifications to generate Gherkin feature files. These feature files are then executed in an environment with high-privilege capabilities.
Ingestion points: User-provided functional specifications (SKILL.md, Step 1).
Boundary markers: The instructions do not define clear boundary markers or sanitization requirements for the input specification to prevent embedded instructions from influencing the generated tests.
Capability inventory: The skill can write files to the project (output_in in ObjectSteps.java), read files from the classpath (load), access environment variables and system properties (_env and _properties proxies), execute shell commands via Maven/Gradle wrappers, and reflectively call Java methods based on Gherkin step text.
Sanitization: The output_in method in ObjectSteps.java includes a path normalization check to prevent directory traversal outside of the resource folder.
[DYNAMIC_EXECUTION]: The framework makes extensive use of dynamic execution patterns to facilitate BDD testing.
Reflection: ObjectSteps.java uses Methods.invokeUnchecked and Methods.findMethod to dynamically map Gherkin text to Java method calls at runtime. This allows the execution of any method available on objects within the test context.
Templating: The skill uses Handlebars (handlebars.compileInline) for dynamic string interpolation and logic within feature files, providing a surface for template injection if inputs are not strictly controlled.
[COMMAND_EXECUTION]: The skill instructions (SKILL.md, Step 4) direct the agent to execute shell commands using project build tools (./mvnw and ./gradlew) to verify the environment and run tests.
[DATA_EXFILTRATION]: ObjectSteps.java provides proxies (_env and _properties) that allow test scenarios to read and set system environment variables and properties. In a CI/CD or developer environment, this capability could be used to expose sensitive credentials if a malicious scenario is generated.

add-cucumber-tests