My Claw Shell

Runs shell commands inside a dedicated tmux session named claw, returning the command output while restricting dangerous commands without confirmation.

This skill will run arbitrary shell commands inside a tmux session named 'claw' and return the pane output — that is its core function and is implemented in handler.js. Before installing, consider: (1) the skill does not declare tmux as a required binary but requires it — ensure tmux is present and you understand where commands will run; (2) its "dangerous command" check is a simple substring heuristic and can miss destructive or exfiltrating commands (e.g., piping remote scripts into sh, use of other destructive tools), so do not rely on it as a safety guarantee; (3) because the skill executes whatever commands the agent or user provides (after approval), any command with network access or file access can exfiltrate data or change the system — prefer running this skill in a restricted/sandboxed environment or with explicit human confirmation for any non-trivial command; (4) if you expect automatic use by the agent, review policy for when the agent is allowed to run commands and consider requiring explicit user confirmation for dangerous patterns. To improve trust: require/declare tmux in the metadata, expand or harden dangerous-command detection, and add explicit guidance about sandboxing or privilege expectations.

Type: OpenClaw Skill Name: my-claw-shell Version: 1.0.0 This skill is suspicious because it provides direct arbitrary shell command execution capabilities via `node:child_process.execSync` in `handler.js`. While it includes a blacklist (`isDangerous` function in `handler.js`) and instructions in `SKILL.md` for the agent to ask for user confirmation for certain commands, these safeguards are weak and easily bypassable. The blacklist is incomplete, and the agent's instructions are susceptible to prompt injection, allowing an attacker to execute arbitrary commands on the host system, leading to potential Remote Code Execution (RCE).