The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides explicit instructions for the agent to download and execute a binary installer from an external CDN (specifically https://cdn.tabulareditor.com/files/cli/latest/te-linux-x64.tar.gz). This allows for the installation and execution of arbitrary code within the agent's runtime environment.
[COMMAND_EXECUTION]: The installation guide directs the agent to modify its persistent shell configuration files (e.g., ~/.bashrc and ~/.zshrc) to add the binary directory to the system PATH. This is a classic persistence mechanism used to ensure tools remain active across sessions.
[DATA_EXFILTRATION]: The skill identifies and describes the exact paths where sensitive service principal tokens and credential caches are stored, such as ~/.te-cli/token-cache.bin and ~/.te-cli/. This information simplifies the discovery and potential exfiltration of credentials.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to fetch configuration and binary data from non-whitelisted external domains, such as cdn.tabulareditor.com.
[PROMPT_INJECTION]: The skill's primary function includes managing 'CustomInstructions' metadata within semantic models. This metadata is specifically intended to override or guide AI behavior, creating a significant attack surface for indirect prompt injection if the model data is sourced from an untrusted or compromised environment.

te-cli