Install OpenClaw skills at the local/system level so they are shared by all agents. Uses the clawhub CLI to fetch and install skills into the global OpenClaw...

Key things to consider before installing: (1) Verify you actually have and trust the `clawhub` CLI on the host; the skill will run it but does not declare that dependency. (2) The code uses string-based shell execution with the provided skill_name; if the agent or a user can supply a crafted skill name, that can lead to command injection (e.g., names containing `;` or `&&`). Prefer a version that spawns the command with arguments (no shell) or validates/whitelists skill names. (3) The skill advertises 'shared/system' installs but the implementation uses the current workspace (OPENCLAW_WORKSPACE or cwd) — expect it to be workspace-scoped unless you inspect/modify the code. (4) Because the agent can invoke this tool autonomously, consider disabling autonomous invocation for this skill or require human confirmation for install actions. (5) If you still want to use it, review or replace the helper scripts so they call `execFile`/spawn with separate args, explicitly require or validate the `clawhub` path, and avoid using undeclared environment variables.

Type: OpenClaw Skill Name: install-shared-skill Version: 0.1.2 The skill bundle contains a critical command injection vulnerability in both tools/install_skill.js and tools/install_shared_skill.js. The code uses child_process.exec to run a shell command constructed with unsanitized input from the skill_name parameter, allowing for arbitrary remote code execution (RCE). While the tool's stated purpose is to facilitate skill installation via the clawhub CLI, the lack of input validation makes it highly dangerous.