Native Run

Execute native commands on the local Windows machine and return their output to OpenClaw for automation and testing purposes.

This skill will run arbitrary shell commands on the gateway machine — a high-risk capability that appears intentional. Before installing: (1) Do not deploy on any machine you don't fully control or trust. (2) Fix the mismatches: update skill.json to point to the actual entry file, and align SKILL.md examples with the JS pattern (or vice versa). (3) Remove the hardcoded token: require a configurable secret via secure environment variable or platform-managed credential, and rotate it. (4) Implement command whitelisting/sanitization and restrict the server to localhost or a secured socket. (5) Consider running the service in an isolated container or VM with limited privileges. If you cannot inspect and remediate the code yourself, treat this package as unsafe to install on production or sensitive systems.

Type: OpenClaw Skill Name: native-run Version: 0.0.1 This skill is highly suspicious due to a critical shell injection vulnerability. The `native_run.py` script executes user-provided commands directly via `subprocess.check_output` with `shell=True`, allowing arbitrary command execution on the host machine. The `native_run_skill.js` passes unsanitized user input from the chat message directly to this vulnerable endpoint. While the skill explicitly states its purpose is to run native commands, the implementation introduces a severe Remote Code Execution (RCE) risk without clear malicious intent like data exfiltration or persistence, thus classifying it as suspicious rather than malicious.