The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches source code and installation scripts from an untrusted GitHub repository (github.com/njerschow/textme).
[REMOTE_CODE_EXECUTION]: Instructs the user to build and run a daemon that facilitates remote command execution on the host machine via inbound iMessages processed by an AI agent.
[COMMAND_EXECUTION]: The daemon enables arbitrary code and shell command execution (including directory navigation and task control) triggered by remote SMS messages.
[PERSISTENCE]: Provides explicit instructions to establish persistent execution using pm2 and macOS launchd services, ensuring the remote-control surface remains active across system reboots.
[INDIRECT_PROMPT_INJECTION]: The skill creates a significant attack surface where untrusted data from inbound iMessages is directly processed by the AI agent, potentially leading to unauthorized actions if the phone-number whitelist is bypassed or compromised.
Ingestion points: Inbound iMessages received via the Sendblue API and polled by the local daemon.
Boundary markers: Relies exclusively on a phone-number whitelist; no internal prompt delimiters or "ignore instructions" warnings are applied to the message content before routing to the agent.
Capability inventory: Full file system access, shell command execution, and directory management via the integrated Claude Code session.
Sanitization: No validation or sanitization of message content is described, allowing the inbound text to drive agent actions directly.

textme