The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents tools such as wsh_create_session, wsh_send_input, and wsh_send_keys which allow the agent to execute shell commands and interact with terminal applications. This is the primary intended function of the skill.
[DATA_EXFILTRATION]: The skill provides instructions for reading the terminal screen and scrollback history via wsh_get_screen and wsh_get_scrollback. This capability allows the agent to access any data displayed in the terminal session.
[PROMPT_INJECTION]: The skill describes a "Fundamental Loop" where the agent is encouraged to read terminal output and decide on next steps based on that content. This creates a surface for indirect prompt injection if the agent reads untrusted data (e.g., file contents or web output) that contains malicious instructions.
Ingestion points: Terminal output retrieved via wsh_get_screen and wsh_get_scrollback (SKILL.md).
Boundary markers: No specific boundary markers or instruction-ignoring wrappers are defined in the bootstrap guidance.
Capability inventory: Full terminal input injection and session management via the wsh_* toolset.
Sanitization: The instructions do not mention sanitizing or escaping terminal output before the agent evaluates it for decision-making.

core-mcp