The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run-and-classify.py uses subprocess.Popen to execute arbitrary commands provided via the command line (argparse.REMAINDER). While the skill's documentation context assumes use for QEMU binaries, the script itself does not restrict the binaries or arguments that can be executed, providing the agent with broad local execution capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze serial console logs generated by external processes (QEMU guests), which are considered untrusted data sources.
Ingestion points: The script scripts/run-and-classify.py reads data from an external log file (args.log), and instructions in SKILL.md direct the agent to analyze and summarize this content in the audit.md file.
Boundary markers: There are no instructions or implementations that use delimiters or specific warnings to ignore instructions that might be embedded within the log content.
Capability inventory: The agent can execute arbitrary commands via the provided runner script and perform file system writes and directory creation within the project workspace.
Sanitization: No filtering, escaping, or sanitization is performed on the log data before it is read by the script or processed by the agent.

qemu-boot-run