The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute dcc-mcp-cli and dcc-cua commands. These commands are used for component management, status verification, and directing UI interactions on the host system.
[EXTERNAL_DOWNLOADS]: The instruction set includes the command dcc-mcp-cli components ensure dcc-cua --yes, which downloads and installs the dcc-cua executable component. The skill's documentation indicates that this tool verifies component integrity using SHA-256 hashes against an official manifest.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to automate and observe external application UIs and web browsers. This creates an attack surface where instructions embedded in the observed content (such as web page text, DOM elements, or window titles) could influence agent behavior.
Ingestion points: Observations of application UI state and browser DOM content processed via ui_control and browser_dom capabilities.
Boundary markers: The skill requires strict binding to specific Process IDs (PID) and Native Window Handles (HWND), and utilizes "action-evidence epochs" to maintain state consistency.
Capability inventory: The agent can execute arbitrary shell commands via Bash (scoped to vendor tools) and perform UI actions like clicks and menu invocations.
Sanitization: The skill contains explicit instructions to avoid exposing sensitive data (credentials, tokens, local usernames) in audit records and prefers semantic references over raw coordinate-based input.

dcc-cua