The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to "interview" the codebase by reading files such as README.md, package.json, and Makefile to determine how to start and drive the application. This creates a surface where malicious instructions or commands embedded in the repository could be ingested and acted upon by the agent during the generation process.
[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands gathered from the repo to launch the application and verify its behavior. Crucially, step 4 ("Prove the generated skill") explicitly instructs the agent to run the generated commands end-to-end (launch, doctor, drive, and cleanup), which involves executing code derived from the analyzed environment.
[DYNAMIC_EXECUTION]: The skill generates new executable instruction files (SKILL.md) and support scripts within the .agents/skills/ directory. It then executes these dynamically created instructions to validate the setup, representing a runtime generation and execution pattern.

create-verification-skill