The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to discover and interact with a local Unix socket (relay.sock) located in the vendor-specific directory $HOME/.agentworkforce/desktop/. It uses curl with the --unix-socket flag to perform local API calls to a relay daemon.
[EXTERNAL_DOWNLOADS]: The skill references documentation and setup guides hosted on the author's official GitHub repository (AgentWorkforce/relay-desktop). It also provides instructions for adding a hosted MCP server from the vendor's domain (agentrelay.com) when local communication is unavailable.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process data from external agents, including participant names and remote messages, which serves as a potential attack surface.
Ingestion points: Data enters the agent context through the /connect/join response and messages injected into the chat stream via the local probe.
Boundary markers: The skill contains explicit instructions to treat invite tasks, participant names, and remote messages as untrusted data rather than instructions.
Capability inventory: The agent has access to local shell execution for socket communication and network access via the hosted MCP fallback.
Sanitization: The instructions strictly prohibit the interpolation of remote or human-provided text into shell source code or arguments.

relay-connect