The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute arbitrary shell scripts, binaries, and commands on the user's host operating system.
Evidence: The skill description and Step 2 explicitly mention running shell scripts, binaries, AppleScript/JXA, and PowerShell commands across macOS, Linux, and Windows.
[PERSISTENCE]: The skill establishes persistence by creating entries in system schedulers that remain active across reboots.
Evidence: Step 2 lists launchd (macOS), cron and systemd (Linux), and Windows Task Scheduler as supported backends for recurring and one-time tasks.
[DYNAMIC_EXECUTION]: The agent is instructed to generate and write executable code to the local file system at runtime.
Evidence: Step 5 states the agent may "Write small helper scripts" to facilitate task execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input and uses it to construct system commands, creating a vulnerability surface for command injection.
Ingestion points: User-provided "Action" strings (messages or commands) parsed in Step 1.
Boundary markers: None identified. There are no instructions to use delimiters or ignore instructions embedded in the user-provided tasks.
Capability inventory: The skill can write files and execute arbitrary shell/system commands through various native schedulers (launchd, cron, systemd, Task Scheduler).
Sanitization: No explicit sanitization, validation, or escaping of shell meta-characters is defined in the instructions.

scheduler