The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents extensive methods to execute shell commands within terminal panes via zellij run and zellij action write-chars. While essential for the tool's purpose, this gives the agent high-privilege control over the host terminal.
[REMOTE_CODE_EXECUTION]: In references/actions.md, the skill describes the zellij action launch-plugin command, which allows loading and executing WebAssembly plugins from arbitrary local file paths (e.g., file:/path/to/plugin.wasm). This facilitates the execution of external binary logic.
[DATA_EXFILTRATION]: The zellij action dump-screen command allows the agent to read the content of terminal panes, including full scrollback history, and write it to a local file. This capability can be used to harvest sensitive data displayed during terminal sessions.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface due to its ability to process terminal output and execute commands based on it. * Ingestion points: Terminal content is read via dump-screen (SKILL.md) and edit-scrollback (references/actions.md). * Boundary markers: Absent; there are no instructions to help the agent distinguish between data and instructions in captured output. * Capability inventory: Arbitrary command execution is available via write-chars and run (SKILL.md), and file-writing is available via dump-screen. * Sanitization: Absent; the skill does not include validation or sanitization steps for the data retrieved from the terminal.

zellij