The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute powerful native macOS utilities such as top, instruments, powermetrics, and ioreg. These are invoked through Python's subprocess.run or shell scripts, which is the primary intended function of the skill.
[COMMAND_EXECUTION]: Documentation suggests the use of sudo to resolve permission issues when running certain tools like powermetrics, which could lead to unintended privilege escalation if mismanaged.
[PROMPT_INJECTION]: The skill represents an indirect prompt injection surface. It ingests data from system command outputs (like process lists from top or system registry info from ioreg) and parses them using regex. If an attacker can influence these inputs (e.g., by naming a malicious process with injection strings), and the agent processes this output without sanitization, it could influence the agent's behavior.
Ingestion points: Outputs from top, vm_stat, ioreg, and instruments (referenced in SKILL.md).
Boundary markers: None implemented; outputs are piped directly to grep or parsed via regex.
Capability inventory: Uses subprocess.run to execute arbitrary system binaries and instruments to launch application paths.
Sanitization: No evidence of escaping or validation of the tool outputs or the application paths provided to instruments.

macos-perf