The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of QEMU commands through a wrapper script scripts/run-and-classify.py. This script uses subprocess.Popen to launch the virtual machine and manage its lifecycle based on output markers. The command is passed as a list of arguments, which is a standard and secure method for process execution within the skill's intended purpose.\n- [DATA_EXFILTRATION]: The skill manages logs and audit files in a local .oh-my-qemu/ directory. It explicitly instructs the agent to configure Git to exclude these directories using the info/exclude file, which is a security best practice to prevent the accidental exposure of build artifacts or logs in version control.\n- [PROMPT_INJECTION]: The skill processes serial log data generated by the QEMU guest environment, representing an indirect prompt injection surface.\n
Ingestion points: The run-and-classify.py script reads from a log file (e.g., uart1.log) that captures the guest serial console output.\n
Boundary markers: The runner uses explicit --success-marker and --failure-marker arguments to delineate expected outcomes, which helps the agent identify state transitions without relying on unparsed guest output.\n
Capability inventory: The script has the capability to execute command-line strings and terminate process groups based on the detected markers.\n
Sanitization: The script performs substring matching on the log content but does not provide active sanitization of the guest output before it is summarized by the agent.

qemu-boot-run