rtk-ai vulnerabilities and threats

rtk filters and compresses command outputs before they reach your LLM context. Prior to 0.42.2, the permission splitter did not conservatively split or reject several shell constructs that Bash treats as command execution boundaries or nested execution. As a result, a command beginning with an allowed prefix such as git could hide a second command behind one of these constructs. rtk rewrite returned exit code 0, causing the Claude hook to emit permissionDecision: "allow". The rewritten command still contained the hidden command, so it ran without the user confirmation or denial that the permission rules were intended to enforce. This vulnerability is fixed in 0.42.2.

rtk filters and compresses command outputs before they reach your LLM context. Prior to 0.32.0, RTK (Rust Token Killer) improperly trusts project-local configuration files. RTK automatically loads .rtk/filters.toml from the working directory with highest priority and without user notification. An attacker can place a malicious filter file in a repository to apply regex-based modifications (e.g., strip_lines_matching) to shell command output before it is shown to the LLM, without any indication that the output has been modified. This allows attackers to selectively suppress or alter command output (including file contents, diffs, and security scan results) without detection, potentially concealing malicious code during AI-assisted development or review. This vulnerability is fixed in 0.32.0.

@rtk-ai/rtk-rewrite transparently rewrites shell commands executed via OpenClaw's exec tool to their RTK equivalents. In 1.0.0, the @rtk-ai/rtk-rewrite OpenClaw plugin passes attacker-controlled input directly into a shell-backed execSync() template string without shell-safe escaping. JSON.stringify() wraps the value in double quotes and escapes inner double-quotes and backslashes, but leaves $() and backtick shell metacharacters untouched. Because execSync delegates execution to /bin/sh -c, the shell expands $(...) substitutions even inside double-quoted strings, causing the injected subcommand to execute before rtk is invoked. An attacker who can influence the exec tool's command parameter (e.g., via an LLM agent prompt or gateway/tool-call input) achieves arbitrary OS command execution with the privileges of the plugin/gateway process.