The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/install.sh script downloads a binary file from a non-trusted GitHub repository (github.com/channprj). The binary is downloaded directly using curl to a local bin directory.
[REMOTE_CODE_EXECUTION]: The installation script sets executable permissions on the downloaded binary and immediately executes it (kmsg --version). This pattern of downloading and running arbitrary binaries from the internet bypasses standard security vetting and can lead to system compromise.
[COMMAND_EXECUTION]: The skill uses osascript to run complex AppleScripts that interact with the system's Accessibility API to control the KakaoTalk UI. It also modifies the user's shell configuration (~/.zshrc) to persist changes to the system PATH environment variable.
[PROMPT_INJECTION]: The skill processes untrusted chat message data which could contain malicious instructions for the agent.
Ingestion points: Reads chat history from KakaoTalk via the kmsg read command and the kmsg_read MCP tool.
Boundary markers: Absent; there are no specific markers or instructions to ignore commands within the message body.
Capability inventory: The skill provides commands to send messages (kmsg send) and perform UI-based automation through AppleScript.
Sanitization: Absent; the skill does not appear to sanitize or validate the content of retrieved messages before presenting them to the agent.

kmsg