The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs local command execution of bundled Python scripts (such as scripts.run_loop, scripts.aggregate_benchmark, and scripts.package_skill) and manages background processes using shell commands like nohup and kill to support its benchmarking and review functionality.\n- [REMOTE_CODE_EXECUTION]: The skill orchestrates subagents (claude-with-access-to-the-skill) to run test cases in parallel. This remote execution is confined to the agent's testing environment for the purpose of validating skill performance.\n- [PROMPT_INJECTION]: The skill facilitates the generation of instructions for other agents, creating a potential surface for indirect prompt injection. However, it explicitly instructs the model to adhere to the 'Principle of Lack of Surprise' and refuse requests intended to facilitate malicious activities or unauthorized access.

skill-creator