The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill suite makes extensive use of sudo to perform system-level audits and modifications in scripts such as firewall-audit.sh, health-audit.sh, kext-audit.sh, and safe-disable-startup.sh. Many operations also necessitate the manual granting of 'Full Disk Access' to the terminal environment to access protected files.
[COMMAND_EXECUTION]: The skill provides the agent with the ability to execute numerous powerful macOS system commands, including launchctl for service management, diskutil for storage operations, pmset for power management, and osascript for interacting with system events.
[DATA_EXFILTRATION]: The skill audits and reads sensitive system data, including privacy permission records from the TCC (Transparency, Consent, Control) databases (TCC.db) and information regarding the user's keychain state (login.keychain-db). It also facilitates the creation of a sysdiagnose bundle, which contains a comprehensive dump of system state and logs.
[INDIRECT_PROMPT_INJECTION]: The skill processes substantial amounts of untrusted data from system logs, configuration files, and command outputs, creating an attack surface where malicious log entries could influence agent behavior.
Ingestion points: System log analysis in panic-triage.sh and health-audit.sh; Plist parsing in startup-audit.sh and safe-disable-startup.sh.
Boundary markers: Most outputs are presented as plain text without explicit delimiters to isolate untrusted data from the agent's instructions.
Capability inventory: The skill possesses capabilities for sudo command execution, service modification via launchctl, and file system manipulation through diskutil and rsync.
Sanitization: A redaction filter in scripts/_lib/common.sh is used to mask personally identifiable information like IP addresses, MAC addresses, and hardware serial numbers.

mac-ops