The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a Python script (scripts/interminai) that acts as a PTY daemon, allowing the agent to execute and interact with arbitrary shell commands (e.g., vim, git, sudo, apt). This is the core functionality of the skill.
[PROMPT_INJECTION]: The skill is designed to supervise other CLI-based AI agents, creating a surface for indirect prompt injection (Category 8).
Ingestion points: External data enters the agent's context through the scripts/interminai output command, which reads the PTY screen buffer (documented in SKILL.md and examples.md).
Boundary markers: The instructions suggest manual verification of the sub-agent's prompt ("verify it is the prompt not some random thought"), but there are no programmatic delimiters or safety instructions to prevent the agent from obeying commands embedded in the sub-agent's output.
Capability inventory: The skill utilizes the Shell tool and can execute any command on the host via os.execvp in the interminai script.
Sanitization: There is no evidence of automated sanitization or filtering of the terminal output before it is processed by the AI agent.
[CREDENTIALS_UNSAFE]: The skill documents interaction with sudo and provides a --password flag to allow users to input credentials directly. Although this is intended as a security feature to keep passwords out of the AI's input logs, the agent still captures the full terminal screen, which could result in data exposure if an application unexpectedly displays sensitive information.

interminai