The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to identify and execute build and startup commands, such as Makefiles or package scripts, discovered in the target repository. It also mandates that the agent autonomously verify the generated verification skill by executing these commands end-to-end. \n- [DYNAMIC_EXECUTION]: The agent is tasked with generating a new project-local skill and immediately executing its instructions, which involves runtime generation and execution of scripted behavior. \n- [INDIRECT_PROMPT_INJECTION]: The skill's operation relies on 'interviewing' untrusted repository data to derive execution logic, creating a surface for injection attacks from malicious codebase content. \n
Ingestion points: Project codebase, documentation, and configuration files are accessed in the interview phase. \n
Boundary markers: No specific delimiters or safety warnings are included in the instructions for processing external repo data. \n
Capability inventory: The agent has capabilities for shell command execution, filesystem writes, and browser automation. \n
Sanitization: The instructions do not require validation or sanitization of the commands or selectors extracted from the repository before they are used in the generated skill.

create-verification-skill