The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes an integration where an AI agent consumes output from the TestBox framework, which presents a surface for indirect prompt injection if the application being tested produces malicious failure messages.
Ingestion points: The skill explicitly recommends the Agent reporter (--reporter=agent), which outputs test failure details in JSON format for agent consumption. It also highlights the 'Ask AI' feature in HTML reports which generates prompts based on failure data.
Boundary markers: There are no instructions or delimiters provided to help the agent distinguish between the framework's report structure and potentially malicious content within the 'message' or 'stack' fields of a failure.
Capability inventory: The agent is instructed to use the BoxLang CLI (./testbox/run) to execute tests, read failure results, and potentially modify the codebase or interact with external AI providers based on those results.
Sanitization: The skill does not mention any sanitization or validation of the test output before it is processed by the AI agent.
[COMMAND_EXECUTION]: The skill provides instructions for executing shell commands, including a feature that allows running arbitrary commands to start web servers.
Evidence: The documentation details the use of the ./testbox/run CLI and specifically the --web-server flag, which executes a user-provided shell string (e.g., --web-server="boxlang-miniserver --port 8080") before running tests.

boxlang-testing