The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell scripts (add-agent-cli.sh and check-agent-cli.sh) to automate the integration process. These scripts perform various system operations including:
Modifying the application's source code using sed and perl -0pi -e to inject the agent routing logic.
Managing file system paths and creating symlinks in ~/.local/bin to install the CLI binary.
Using lsof and kill to manage test instances of the application during automated checks.
[DYNAMIC_EXECUTION]: The skill implements a custom RPC (Remote Procedure Call) mechanism where the CLI and MCP server send JSON-formatted commands over a Unix domain socket to the running macOS application. The application then dynamically routes these requests to pre-defined Swift closures. While this enables runtime command execution, the available commands are statically defined in the source code (AgentCommands.swift).
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted application data (such as item titles, notes, or status information).
Ingestion points: Data enters the agent's context through CLI output and MCP tool results (e.g., items list, status).
Boundary markers: The skill documentation recommends using the --json flag to ensure structured data output, which helps the agent distinguish between data and instructions.
Capability inventory: The skill provides capabilities to read, write, and delete application data, as well as control the application's UI state.
Sanitization: The implementation relies on standard JSON serialization and does not explicitly sanitize natural language content within the data fields.

agent-cli