output-compression

Warn

Audited by Gen Agent Trust Hub on Apr 7, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and install third-party software from external sources.
  • It recommends installing rtk and tokf using package managers like Homebrew and Cargo.
  • Specifically, it references a third-party Homebrew tap mpecan/tokf/tokf which is not associated with a verified or trusted organization.
  • [COMMAND_EXECUTION]: The skill requires the execution of commands that modify the shell environment and agent behavior.
  • The commands rtk init --global and tokf hook install set up "PreToolUse" hooks.
  • These hooks are designed to intercept and transparently rewrite Bash commands before they are processed by the agent, which is a form of environment persistence and command modification.
  • [REMOTE_CODE_EXECUTION]: The tokf tool supports project-level customization through files stored in a repository's .tokf/filters/ directory.
  • These configuration files can include Luau (Lua) scripts for processing command output.
  • If an agent is used within a repository containing a malicious .tokf configuration, it could lead to the execution of arbitrary script logic during the tool's operation.
  • [PROMPT_INJECTION]: The mechanism of filtering and rewriting CLI output creates a surface for indirect prompt injection.
  • Malicious actors could commit custom filters to a repository that selectively hide, modify, or inject content into the command output before it reaches the AI's context window.
  • This could lead to the agent making decisions based on misrepresented or incomplete data without the user's knowledge.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 7, 2026, 03:50 PM