The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill establishes a remote command bridge by installing a daemon that routes inbound iMessages to a Claude Code session, enabling the remote execution of code and filesystem operations from a mobile device.
[EXTERNAL_DOWNLOADS]: Fetches and installs software from a community GitHub repository (njerschow/textme) and installs external dependencies via npm to facilitate the bridge between external messages and the local system.
[PERSISTENCE]: Configures the remote access daemon to remain active across user sessions and system reboots using pm2 (pm2 startup, pm2 save) and macOS launchd service files.
[PRIVILEGE_ESCALATION]: Includes instructions for registering the daemon as a system-level service via pm2 startup and a custom installation script (./scripts/install-launchd.sh), which can grant the daemon persistent elevated status on the host.
[INDIRECT_PROMPT_INJECTION]: The skill creates a significant attack surface by ingesting untrusted data from an external messaging service and passing it to an AI agent with terminal access.
Ingestion points: Inbound iMessages received via the Sendblue API (SKILL.md).
Boundary markers: The system relies exclusively on a phone-number whitelist for authorization to execute commands (SKILL.md).
Capability inventory: Full filesystem access, directory navigation (cd), and terminal command execution via Claude Code (SKILL.md).
Sanitization: The instructions do not specify any input validation or sanitization for incoming message content before it is processed by the execution environment.
[COMMAND_EXECUTION]: The daemon directly parses and runs administrative commands on the host machine, including directory changes (cd), process termination (stop), and state resets.

textme