The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The skill is designed to execute arbitrary shell commands via .tape files (using Type and Enter commands). While this is the primary purpose of the tool, it allows the agent to execute any command in the local environment, including those that might be malicious if the agent is compromised or misled.
DATA_EXFILTRATION (MEDIUM): The vhs publish command uploads recordings to vhs.charm.sh, a public hosting service. This represents a significant exfiltration vector if the terminal session contains sensitive information like API keys, credentials, or private file contents.
EXTERNAL_DOWNLOADS (LOW): The skill references and encourages the installation of external tools from github.com/charmbracelet/vhs and github.com/marcfargas/holdpty. Neither 'charmbracelet' nor 'marcfargas' are included in the provided list of Trusted GitHub Organizations.
INDIRECT_PROMPT_INJECTION (LOW): The skill creates a surface where untrusted data could be used to generate .tape files, leading to the execution of unintended commands.
Ingestion points: Data used to populate .tape files (e.g., tool output, user input).
Boundary markers: Absent; there are no instructions to sanitize or delimit untrusted content within the tape syntax.
Capability inventory: Shell command execution, network upload (publish), and background process manipulation via holdpty.
Sanitization: None; the skill does not provide methods for escaping or validating the content typed into the terminal.

vhs