The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses sudo extensively for system management tasks, such as modifying service states with launchctl, auditing the packet filter with pfctl, and executing binary diagnostics with sysdiagnose. It also utilizes osascript to run AppleScript for auditing user login items and system events.\n- [DATA_EXFILTRATION]: The diagnostics involve reading sensitive system databases, including the TCC (Transparency, Consent, Control) database to audit application permissions and the login Keychain to verify certificate and lock status. These actions provide visibility into highly sensitive user security configurations.\n- [EXTERNAL_DOWNLOADS]: The skill documentation and the brew-health.sh script reference the Homebrew package manager, a well-known third-party service, and recommend it for installing diagnostic utilities like smartmontools and gddrescue.\n- [PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests untrusted data from system logs and file metadata without implementing boundary markers or sanitization.\n
Ingestion points: System log data via log show, file names from directories like /Library/LaunchAgents, and records from the TCC database are read and processed by scripts such as health-audit.sh and startup-audit.sh.\n
Boundary markers: No explicit delimiters or instructions are used to distinguish ingested system data from agent instructions.\n
Capability inventory: The skill is granted Bash, Read, and Write tools and operates with sudo privileges for several utilities.\n
Sanitization: There is no evidence of sanitization, escaping, or filtering of ingested log or metadata content before it is passed to the agent context.

mac-ops